ES2000_MathRock_Diffusion.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/ethansmith2000/MathRock-Diffusion/blob/main/ES2000_MathRock_Diffusion.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TitleTop"
      },
      "source": [
        "# ES2000's MathRock Diffusion BETA\n",
        "adapted from DiscoDiffusion v5.61\n",
        "namesake inspired by my favorite genre of music, and ProgRockDiffusion\n",
        "\n",
        "a continuous work in progress notebook, with a variety of adaptations and mods for disco\n",
        "\n",
        "join our discord server: https://discord.gg/bsstKzXHt7\n",
        "\n",
        "questions, feedback, etc.\n",
        "[linktree and contact](https://linktr.ee/ethansmith2000)\n",
        "\n",
        "\n",
        "------------------------\n",
        "\n",
        "Disco Diffusion - http://discodiffusion.com/ , https://github.com/alembics/disco-diffusion\n",
        "\n",
        "In case of confusion, Disco is the name of this notebook edit. The diffusion model in use is Katherine Crowson's fine-tuned 512x512 model\n",
        "\n",
        "For issues, join the [Disco Diffusion Discord](https://discord.gg/msEZBy4HxA) or message us on twitter at [@somnai_dreams](https://twitter.com/somnai_dreams) or [@gandamu_ml](https://twitter.com/gandamu_ml)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CreditsChTop"
      },
      "source": [
        "# Credits & Changelog ⬇️"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Credits"
      },
      "source": [
        "#### Credits\n",
        "\n",
        "Original notebook by Katherine Crowson (https://github.com/crowsonkb, https://twitter.com/RiversHaveWings). It uses either OpenAI's 256x256 unconditional ImageNet or Katherine Crowson's fine-tuned 512x512 diffusion model (https://github.com/openai/guided-diffusion), together with CLIP (https://github.com/openai/CLIP) to connect text prompts with images.\n",
        "\n",
        "Modified by Daniel Russell (https://github.com/russelldc, https://twitter.com/danielrussruss) to include (hopefully) optimal params for quick generations in 15-100 timesteps rather than 1000, as well as more robust augmentations.\n",
        "\n",
        "Further improvements from Dango233 and nshepperd helped improve the quality of diffusion in general, and especially so for shorter runs like this notebook aims to achieve.\n",
        "\n",
        "Vark added code to load in multiple Clip models at once, which all prompts are evaluated against, which may greatly improve accuracy.\n",
        "\n",
        "The latest zoom, pan, rotation, and keyframes features were taken from Chigozie Nri's VQGAN Zoom Notebook (https://github.com/chigozienri, https://twitter.com/chigozienri)\n",
        "\n",
        "Advanced DangoCutn Cutout method is also from Dango223.\n",
        "\n",
        "--\n",
        "\n",
        "Disco:\n",
        "\n",
        "Somnai (https://twitter.com/Somnai_dreams) added Diffusion Animation techniques, QoL improvements and various implementations of tech and techniques, mostly listed in the changelog below.\n",
        "\n",
        "3D animation implementation added by Adam Letts (https://twitter.com/gandamu_ml) in collaboration with Somnai. Creation of disco.py and ongoing maintenance.\n",
        "\n",
        "Turbo feature by Chris Allen (https://twitter.com/zippy731)\n",
        "\n",
        "Improvements to ability to run on local systems, Windows support, and dependency installation by HostsServer (https://twitter.com/HostsServer)\n",
        "\n",
        "VR Mode by Tom Mason (https://twitter.com/nin_artificial)\n",
        "\n",
        "Horizontal and Vertical symmetry functionality by nshepperd. Symmetry transformation_steps by huemin (https://twitter.com/huemin_art). Symmetry integration into Disco Diffusion by Dmitrii Tochilkin (https://twitter.com/cut_pow).\n",
        "\n",
        "Warp and custom model support by Alex Spirin (https://twitter.com/devdef).\n",
        "\n",
        "Pixel Art Diffusion, Watercolor Diffusion, and Pulp SciFi Diffusion models from KaliYuga (https://twitter.com/KaliYuga_ai). Follow KaliYuga's Twitter for the latest models and for notebooks with specialized settings.\n",
        "\n",
        "Integration of OpenCLIP models and initiation of integration of KaliYuga models by Palmweaver / Chris Scalf (https://twitter.com/ChrisScalf11)\n",
        "\n",
        "Integrated portrait_generator_v001 from Felipe3DArtist (https://twitter.com/Felipe3DArtist)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LicenseTop"
      },
      "source": [
        "#### License"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "License"
      },
      "source": [
        "Licensed under the MIT License\n",
        "\n",
        "Copyright (c) 2021 Katherine Crowson \n",
        "\n",
        "Permission is hereby granted, free of charge, to any person obtaining a copy\n",
        "of this software and associated documentation files (the \"Software\"), to deal\n",
        "in the Software without restriction, including without limitation the rights\n",
        "to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n",
        "copies of the Software, and to permit persons to whom the Software is\n",
        "furnished to do so, subject to the following conditions:\n",
        "\n",
        "The above copyright notice and this permission notice shall be included in\n",
        "all copies or substantial portions of the Software.\n",
        "\n",
        "THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n",
        "IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n",
        "FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n",
        "AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n",
        "LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n",
        "OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n",
        "THE SOFTWARE.\n",
        "\n",
        "--\n",
        "\n",
        "MIT License\n",
        "\n",
        "Copyright (c) 2019 Intel ISL (Intel Intelligent Systems Lab)\n",
        "\n",
        "Permission is hereby granted, free of charge, to any person obtaining a copy\n",
        "of this software and associated documentation files (the \"Software\"), to deal\n",
        "in the Software without restriction, including without limitation the rights\n",
        "to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n",
        "copies of the Software, and to permit persons to whom the Software is\n",
        "furnished to do so, subject to the following conditions:\n",
        "\n",
        "The above copyright notice and this permission notice shall be included in all\n",
        "copies or substantial portions of the Software.\n",
        "\n",
        "THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n",
        "IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n",
        "FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n",
        "AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n",
        "LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n",
        "OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\n",
        "SOFTWARE.\n",
        "\n",
        "--\n",
        "\n",
        "Licensed under the MIT License\n",
        "\n",
        "Copyright (c) 2021 Maxwell Ingham\n",
        "\n",
        "Copyright (c) 2022 Adam Letts \n",
        "\n",
        "Copyright (c) 2022 Alex Spirin\n",
        "\n",
        "Permission is hereby granted, free of charge, to any person obtaining a copy\n",
        "of this software and associated documentation files (the \"Software\"), to deal\n",
        "in the Software without restriction, including without limitation the rights\n",
        "to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n",
        "copies of the Software, and to permit persons to whom the Software is\n",
        "furnished to do so, subject to the following conditions:\n",
        "\n",
        "The above copyright notice and this permission notice shall be included in\n",
        "all copies or substantial portions of the Software.\n",
        "\n",
        "THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n",
        "IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n",
        "FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n",
        "AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n",
        "LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n",
        "OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN\n",
        "THE SOFTWARE.\n",
        "\n",
        "--\n",
        "flow-related - https://github.com/NVIDIA/flownet2-pytorch/blob/master/LICENSE\n",
        "--\n",
        "Copyright 2017 NVIDIA CORPORATION\n",
        "\n",
        "Licensed under the Apache License, Version 2.0 (the \"License\");\n",
        "you may not use this file except in compliance with the License.\n",
        "You may obtain a copy of the License at\n",
        "\n",
        "    http://www.apache.org/licenses/LICENSE-2.0\n",
        "\n",
        "Unless required by applicable law or agreed to in writing, software\n",
        "distributed under the License is distributed on an \"AS IS\" BASIS,\n",
        "WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
        "See the License for the specific language governing permissions and\n",
        "limitations under the License."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ChangelogTop"
      },
      "source": [
        "#### Changelog"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "Changelog"
      },
      "source": [
        "#@title <- View Changelog\n",
        "skip_for_run_all = True #@param {type: 'boolean'}\n",
        "\n",
        "if skip_for_run_all == False:\n",
        "  print(\n",
        "      '''\n",
        "  v1 Update: Oct 29th 2021 - Somnai\n",
        "\n",
        "      QoL improvements added by Somnai (@somnai_dreams), including user friendly UI, settings+prompt saving and improved google drive folder organization.\n",
        "\n",
        "  v1.1 Update: Nov 13th 2021 - Somnai\n",
        "\n",
        "      Now includes sizing options, intermediate saves and fixed image prompts and perlin inits. unexposed batch option since it doesn't work\n",
        "\n",
        "  v2 Update: Nov 22nd 2021 - Somnai\n",
        "\n",
        "      Initial addition of Katherine Crowson's Secondary Model Method (https://colab.research.google.com/drive/1mpkrhOjoyzPeSWy2r7T8EYRaU7amYOOi#scrollTo=X5gODNAMEUCR)\n",
        "\n",
        "      Noticed settings were saving with the wrong name so corrected it. Let me know if you preferred the old scheme.\n",
        "\n",
        "  v3 Update: Dec 24th 2021 - Somnai\n",
        "\n",
        "      Implemented Dango's advanced cutout method\n",
        "\n",
        "      Added SLIP models, thanks to NeuralDivergent\n",
        "\n",
        "      Fixed issue with NaNs resulting in black images, with massive help and testing from @Softology\n",
        "\n",
        "      Perlin now changes properly within batches (not sure where this perlin_regen code came from originally, but thank you)\n",
        "\n",
        "  v4 Update: Jan 2022 - Somnai\n",
        "\n",
        "      Implemented Diffusion Zooming\n",
        "\n",
        "      Added Chigozie keyframing\n",
        "\n",
        "      Made a bunch of edits to processes\n",
        "  \n",
        "  v4.1 Update: Jan 14th 2022 - Somnai\n",
        "\n",
        "      Added video input mode\n",
        "\n",
        "      Added license that somehow went missing\n",
        "\n",
        "      Added improved prompt keyframing, fixed image_prompts and multiple prompts\n",
        "\n",
        "      Improved UI\n",
        "\n",
        "      Significant under the hood cleanup and improvement\n",
        "\n",
        "      Refined defaults for each mode\n",
        "\n",
        "      Added latent-diffusion SuperRes for sharpening\n",
        "\n",
        "      Added resume run mode\n",
        "\n",
        "  v4.9 Update: Feb 5th 2022 - gandamu / Adam Letts\n",
        "\n",
        "      Added 3D\n",
        "\n",
        "      Added brightness corrections to prevent animation from steadily going dark over time\n",
        "\n",
        "  v4.91 Update: Feb 19th 2022 - gandamu / Adam Letts\n",
        "\n",
        "      Cleaned up 3D implementation and made associated args accessible via Colab UI elements\n",
        "\n",
        "  v4.92 Update: Feb 20th 2022 - gandamu / Adam Letts\n",
        "\n",
        "      Separated transform code\n",
        "\n",
        "  v5.01 Update: Mar 10th 2022 - gandamu / Adam Letts\n",
        "\n",
        "      IPython magic commands replaced by Python code\n",
        "\n",
        "  v5.1 Update: Mar 30th 2022 - zippy / Chris Allen and gandamu / Adam Letts\n",
        "\n",
        "      Integrated Turbo+Smooth features from Disco Diffusion Turbo -- just the implementation, without its defaults.\n",
        "\n",
        "      Implemented resume of turbo animations in such a way that it's now possible to resume from different batch folders and batch numbers.\n",
        "\n",
        "      3D rotation parameter units are now degrees (rather than radians)\n",
        "\n",
        "      Corrected name collision in sampling_mode (now diffusion_sampling_mode for plms/ddim, and sampling_mode for 3D transform sampling)\n",
        "\n",
        "      Added video_init_seed_continuity option to make init video animations more continuous\n",
        "\n",
        "  v5.1 Update: Apr 4th 2022 - MSFTserver aka HostsServer\n",
        "\n",
        "      Removed pytorch3d from needing to be compiled with a lite version specifically made for Disco Diffusion\n",
        "\n",
        "      Remove Super Resolution\n",
        "\n",
        "      Remove SLIP Models\n",
        "\n",
        "      Update for crossplatform support\n",
        "\n",
        "  v5.2 Update: Apr 10th 2022 - nin_artificial / Tom Mason\n",
        "\n",
        "      VR Mode\n",
        "\n",
        "  v5.3 Update: Jun 10th 2022 - nshepperd, huemin, cut_pow / Dmitrii Tochilkin\n",
        "\n",
        "      Horizontal and Vertical symmetry\n",
        "\n",
        "      Addition of ViT-L/14@336px model (requires high VRAM)\n",
        "\n",
        "  v5.4 Update: Jun 14th 2022 - devdef / Alex Spirin, Alex's Warp changes integrated into DD main by gandamu / Adam Letts\n",
        "\n",
        "      Warp mode - for smooth/continuous video input results leveraging optical flow estimation and frame blending\n",
        "\n",
        "      Custom models support\n",
        "\n",
        "  v5.5 Update: Jul 11th 2022 - Palmweaver / Chris Scalf, KaliYuga_ai, further DD integration by gandamu / Adam Letts\n",
        "\n",
        "      OpenCLIP models integration\n",
        "\n",
        "      Pixel Art Diffusion, Watercolor Diffusion, and Pulp SciFi Diffusion models\n",
        "\n",
        "      cut_ic_pow scheduling\n",
        "\n",
        "  v5.6 Update: Jul 13th 2022 - Felipe3DArtist integration by gandamu / Adam Letts\n",
        "\n",
        "      portrait_generator_v001 diffusion model integrated\n",
        "\n",
        "  v5.61 Update: Aug 21st 2022 - gandamu / Adam Letts\n",
        "\n",
        "      Correct progress bars issue caused by recent Google Colab environment changes\n",
        "      Adjust 512x512 diffusion model download URI priority due to issues with the previous primary source\n",
        "    '''\n",
        "  )"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TutorialTop"
      },
      "source": [
        "# Tutorial"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DiffusionSet"
      },
      "source": [
        "**Diffusion settings**\n",
        "---\n",
        "A basic rundown on the settings and some suggested values, more can be found in my guide at https://www.notion.so/A-Traveler-s-Guide-to-the-Latent-Space-85efba7e5e6a40e5bd3cae980f30235f#9c774b07e7b34b3d833a074569afb538\n",
        "\n",
        "We also encourage users to join the [Disco Diffusion User Discord](https://discord.gg/XGZrFFCRfN) to learn from the active user community.\n",
        "\n",
        "This section below is outdated as of v2\n",
        "\n",
        "the default values ranges are meant to be a good place to start, feel free to experiment beyond\n",
        "\n",
        "Setting | Description | Default\n",
        "--- | --- | ---\n",
        "**Your vision:**\n",
        "`text_prompts` | A description of what you'd like the machine to generate. | N/A\n",
        "`image_prompts` | upload an image to guide generation (really does not work well) | N/A\n",
        "**Image quality:**\n",
        "`clip_guidance_scale`  | Controls how much the image should look like the prompt. | 5000-30.000\n",
        "`tv_scale` | Controls the smoothness of the final output, higher = more smooth, only works with secondary model turned off | 5,000 - 50,000\n",
        "`range_scale` | Controls contrast/exposure, (currently does not work) | 10,000 - 50,000\n",
        "`sat_scale` | Controls saturation, higher = less saturation, does not work unless secondary model is turned off, From nshepperd's JAX notebook. | 5k-50k\n",
        "`cutn_batches` | batches of cuts, higher generally means more quality, but doubling batches will roughly double time of a run | 2-8\n",
        "`steps` | amount of steps to complete an image generation, more tends to be better but not always, it is best to do a number that divides evenly into 1000 based on how model was trained | 166, 200, 250, 333, 500\n",
        "`symmetry loss` | coaxes the run into making symmetrical outputs higher scale and higher number for switch steps = more symmetry | 160,000\n",
        "`symm switch percent` | % into the run when you want to turn symmetrical guidance off | ~ 30-50%\n",
        "`symmetry transformation` | more aggressive/absolute than symmetry loss, works like mirroring in a photo editor, can do horizontal or vertical\n",
        "`transformation percent` | percent into the run when you want the mirroring to execute, higher values will give more absolute symmetry, too early will result in little effect | [0.4]\n",
        "`clamp_max` | can be thought of a master knob for much of what goes on, higher values = higher saturation, exposure, detail. too much can cause issues, more details on guide | 0.05-0.10+\n",
        "**Init settings:**\n",
        "`init_image` | think of this as a template for the program to work on, feed it an image to start from | None\n",
        "`init_scale` | higher values attempt to preserve more of the init_image you give it | 0\n",
        "`skip_steps` | necessary if using an init_image, ~50% of total steps is suggested, lower values give more ai influence, higher values make it look more like the original image | 100-150 skip on a 250 step run\n",
        "`perlin_init` | Option to start with random perlin noise, must be using skip steps to take effect | False\n",
        "`perlin_mode` | ('gray', 'color') | 'mixed'\n",
        "`resolution` | size of the image you will generate, see the guide for more details | 1024x768\n",
        "`cut_ic_pow` | can be thought to control balance vs detail, higher values = more detail, but also more prone to repetition + other distortions, see guide for more | 1-10\n",
        "`clip preceptors (clip_models)` | Models of CLIP to load. Typically the more, the better but they all come at a hefty VRAM cost. see guide for details | ViT-B/32, ViT-B/16, RN50x4 (VITL14,RN50x4,RN50x16 are all good options if memory allows) \n",
        "**Advanced:**\n",
        "`seed`  | a random seed used to guide a generation, think of seeds for minecraft worlds, using a fixed number with the same prompt will generally give you similar if not identical results, can put a number or \"random_seed\" | random_seed\n",
        "`fuzzy_prompt` | Controls whether to add multiple noisy prompts to the prompt losses | False\n",
        "`rand_mag` | Controls the magnitude of the random noise | 0.1\n",
        "`eta` | ratio of DDIM/DDPM noise, see guide for examples, effect is difficult to put words to | 0.8\n",
        "`cut_overview` | # of cuts that see the whole image, good for early structure, this is scheduled to change throughout run, see guide for details |\n",
        "`cut_innercut` | # of cuts traditional cuts, vary in size depending on cut_pow, good for detail, this is scheduled to change throughout run, see guide for details |\n",
        "`cut_icgray` | % of innercuts that will see the image in grayscale, this may be useful in earlier parts of run, extra contrast may help it find structure, this is scheduled to change throughout run, see guide for details |\n",
        "`ov_gray` | % of overview cuts that will see in grayscale, this may be useful in earlier parts of run, extra contrast may help it find structure, this is scheduled to change throughout run, see guide for details | [0.1]*1000\n",
        "`diffusion sampling mode` | method of noise sampling method, PLMS and IPLMS are often considered better in studies, but not well adapted for our program, mixed modes begin with others then switch back to DDIM/DDPM as a workaround | DDIM\n",
        "`switch percent` | % into run when to switch modes | 30-50%\n",
        "**Others**\n",
        "`skip_augs` | Controls whether to skip torchvision augmentations, hidden in code due to low usage of this function | False\n",
        "`randomize_class` | Controls whether the imagenet class is randomly changed each iteration, hidden in code due to low usage of this function | True\n",
        "`clip_denoised` | Determines whether CLIP discriminates a noisy or denoised image, hidden in code due to low usage of this function | False\n",
        "`clamp_grad` | Experimental: Using adaptive clip grad in the cond_fn, hidden in code to low usage, putting clamp_max at 1.0 should yield the same effect as turning it off | True"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SetupTop"
      },
      "source": [
        "# 1. Set Up"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# uncomment these for a neater UI if using a jupyter notebook\n",
        "\n",
        "# !pip install jupyter_contrib_nbextensions\n",
        "# !jupyter contrib nbextension install --user\n",
        "# !jupyter nbextensions_configurator enable --user\n",
        "# !jupyter nbextension enable codefolding/main\n",
        "# !jupyter nbextension enable toc2/main"
      ],
      "metadata": {
        "id": "9S1x-LzLFDTZ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%%capture\n",
        "\n",
        "#@title libraries for runpod/vast \n",
        "print(\"Installing libs...\")\n",
        "!apt-get update\n",
        "!apt install -y git curl ffmpeg libsm6 libxext6\n",
        "!pip install opencv-python pandas matplotlib regex\n",
        "!export PATH=/usr/bin/ffmpeg:$PATH\n",
        "!echo $PATH\n"
      ],
      "metadata": {
        "id": "Oy0ppgXYC6HE",
        "cellView": "form"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "CheckGPU"
      },
      "source": [
        "#@title 1.1 Check GPU Status\n",
        "import subprocess\n",
        "simple_nvidia_smi_display = False#@param {type:\"boolean\"}\n",
        "if simple_nvidia_smi_display:\n",
        "    #!nvidia-smi\n",
        "    nvidiasmi_output = subprocess.run(['nvidia-smi', '-L'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(nvidiasmi_output)\n",
        "else:\n",
        "    #!nvidia-smi -i 0 -e 0\n",
        "    nvidiasmi_output = subprocess.run(['nvidia-smi'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(nvidiasmi_output)\n",
        "    nvidiasmi_ecc_note = subprocess.run(['nvidia-smi', '-i', '0', '-e', '0'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(nvidiasmi_ecc_note)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "PrepFolders"
      },
      "source": [
        "#@title 1.2 Prepare Folders\n",
        "import subprocess, os, sys, ipykernel\n",
        "\n",
        "def gitclone(url, targetdir=None):\n",
        "    if targetdir:\n",
        "        res = subprocess.run(['git', 'clone', url, targetdir], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    else:\n",
        "        res = subprocess.run(['git', 'clone', url], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "def pipi(modulestr):\n",
        "    res = subprocess.run(['pip', 'install', modulestr], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "def pipie(modulestr):\n",
        "    res = subprocess.run(['git', 'install', '-e', modulestr], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "def wget(url, outputdir):\n",
        "    res = subprocess.run(['wget', url, '-P', f'{outputdir}'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(res)\n",
        "\n",
        "try:\n",
        "    from google.colab import drive\n",
        "    print(\"Google Colab detected. Using Google Drive.\")\n",
        "    is_colab = True\n",
        "    #@markdown If you connect your Google Drive, you can save the final image of each run on your drive.\n",
        "    google_drive = True #@param {type:\"boolean\"}\n",
        "    #@markdown Click here if you'd like to save the diffusion model checkpoint file to (and/or load from) your Google Drive:\n",
        "    save_models_to_google_drive = True #@param {type:\"boolean\"}\n",
        "    #@markdown Click here if you'd like to save the CLIP models to (and/or load from) your Google Drive:\n",
        "    save_clipmodels_to_google_drive = False #@param {type:\"boolean\"}\n",
        "    print(\"Downgrading ipywidgets to latest 7.x in order to enable custom widget manager (for tqdm progress bars)\")\n",
        "    multipip_res = subprocess.run(['pip', 'install', 'ipywidgets>=7,<8'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    print(multipip_res)\n",
        "    from google.colab import output\n",
        "    output.enable_custom_widget_manager()\n",
        "except:\n",
        "    is_colab = False\n",
        "    google_drive = False\n",
        "    save_models_to_google_drive = False\n",
        "    save_clipmodels_to_google_drive = False\n",
        "    print(\"Google Colab not detected.\")\n",
        "    multipip_res = subprocess.run(['pip', 'install', 'ipywidgets>=7,<8'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "\n",
        "if is_colab:\n",
        "    if google_drive is True:\n",
        "        drive.mount('/content/drive')\n",
        "        root_path = '/content/drive/MyDrive/AI/Disco_Diffusion'\n",
        "    else:\n",
        "        root_path = '/content'\n",
        "else:\n",
        "    root_path = os.getcwd()\n",
        "\n",
        "import os\n",
        "def createPath(filepath):\n",
        "    os.makedirs(filepath, exist_ok=True)\n",
        "\n",
        "initDirPath = f'{root_path}/init_images'\n",
        "createPath(initDirPath)\n",
        "outDirPath = f'{root_path}/images_out'\n",
        "createPath(outDirPath)\n",
        "\n",
        "if is_colab:\n",
        "    if google_drive and not save_models_to_google_drive or not google_drive:\n",
        "        model_path = '/content/models'\n",
        "        createPath(model_path)\n",
        "    if google_drive and save_models_to_google_drive:\n",
        "        model_path = f'{root_path}/models'\n",
        "        createPath(model_path)\n",
        "else:\n",
        "    model_path = f'{root_path}/models'\n",
        "    createPath(model_path)\n",
        "\n",
        "# libraries = f'{root_path}/libraries'\n",
        "# createPath(libraries)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "InstallDeps",
        "cellView": "form"
      },
      "source": [
        "%%capture\n",
        "\n",
        "#@title ### 1.3 Install, import dependencies and set up runtime devices\n",
        "\n",
        "import pathlib, shutil, os, sys\n",
        "\n",
        "# There are some reports that with a T4 or V100 on Colab, downgrading to a previous version of PyTorch may be necessary.\n",
        "# .. but there are also reports that downgrading breaks them!  If you're facing issues, you may want to try uncommenting and running this code.\n",
        "# nvidiasmi_output = subprocess.run(['nvidia-smi'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "# cards_requiring_downgrade = [\"Tesla T4\", \"V100\"]\n",
        "# if is_colab:\n",
        "#     if any(cardstr in nvidiasmi_output for cardstr in cards_requiring_downgrade):\n",
        "#         print(\"Downgrading pytorch. This can take a couple minutes ...\")\n",
        "#         downgrade_pytorch_result = subprocess.run(['pip', 'install', 'torch==1.10.2', 'torchvision==0.11.3', '-q'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "#         print(\"pytorch downgraded.\")\n",
        "\n",
        "#@markdown Check this if you want to use CPU\n",
        "useCPU = False #@param {type:\"boolean\"}\n",
        "\n",
        "if not is_colab:\n",
        "    # If running locally, there's a good chance your env will need this in order to not crash upon np.matmul() or similar operations.\n",
        "    os.environ['KMP_DUPLICATE_LIB_OK']='TRUE'\n",
        "\n",
        "PROJECT_DIR = os.path.abspath(os.getcwd())\n",
        "USE_ADABINS = True\n",
        "\n",
        "if is_colab:\n",
        "    if not google_drive:\n",
        "        root_path = f'/content'\n",
        "        model_path = '/content/models' \n",
        "else:\n",
        "    root_path = os.getcwd()\n",
        "    model_path = f'{root_path}/models'\n",
        "\n",
        "multipip_res = subprocess.run(['pip', 'install', 'lpips', 'datetime', 'timm', 'ftfy', 'einops', 'pytorch-lightning', 'omegaconf'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "print(multipip_res)\n",
        "\n",
        "if is_colab:\n",
        "    subprocess.run(['apt', 'install', 'imagemagick'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "\n",
        "try:\n",
        "    from CLIP import clip\n",
        "except:\n",
        "    if not os.path.exists(\"CLIP\"):\n",
        "        gitclone(\"https://github.com/openai/CLIP\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/CLIP')\n",
        "\n",
        "try:\n",
        "    import open_clip\n",
        "except:\n",
        "    if not os.path.exists(\"open_clip/src\"):\n",
        "        gitclone(\"https://github.com/mlfoundations/open_clip.git\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/open_clip/src')\n",
        "    import open_clip\n",
        "\n",
        "try:\n",
        "    from guided_diffusion.script_util import create_model_and_diffusion\n",
        "except:\n",
        "    if not os.path.exists(\"guided-diffusion\"):\n",
        "        gitclone(\"https://github.com/ethansmith2000/guided-diffusion.git\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/guided-diffusion')\n",
        "\n",
        "try:\n",
        "    from resize_right import resize\n",
        "except:\n",
        "    if not os.path.exists(\"ResizeRight\"):\n",
        "        gitclone(\"https://github.com/assafshocher/ResizeRight.git\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/ResizeRight')\n",
        "\n",
        "try:\n",
        "    import py3d_tools\n",
        "except:\n",
        "    if not os.path.exists('pytorch3d-lite'):\n",
        "        gitclone(\"https://github.com/MSFTserver/pytorch3d-lite.git\")\n",
        "    sys.path.append(f'{PROJECT_DIR}/pytorch3d-lite')\n",
        "\n",
        "try:\n",
        "    from midas.dpt_depth import DPTDepthModel\n",
        "except:\n",
        "    if not os.path.exists('MiDaS'):\n",
        "        gitclone(\"https://github.com/isl-org/MiDaS.git\")\n",
        "    if not os.path.exists('MiDaS/midas_utils.py'):\n",
        "        shutil.move('MiDaS/utils.py', 'MiDaS/midas_utils.py')\n",
        "    if not os.path.exists(f'{model_path}/dpt_large-midas-2f21e586.pt'):\n",
        "        wget(\"https://github.com/intel-isl/DPT/releases/download/1_0/dpt_large-midas-2f21e586.pt\", model_path)\n",
        "    sys.path.append(f'{PROJECT_DIR}/MiDaS')\n",
        "\n",
        "try:\n",
        "    sys.path.append(PROJECT_DIR)\n",
        "    import disco_xform_utils as dxf\n",
        "except:\n",
        "    if not os.path.exists(\"disco-diffusion\"):\n",
        "        gitclone(\"https://github.com/alembics/disco-diffusion.git\")\n",
        "    if not os.path.exists('disco_xform_utils.py'):\n",
        "        shutil.move('disco-diffusion/disco_xform_utils.py', 'disco_xform_utils.py')\n",
        "    sys.path.append(PROJECT_DIR)\n",
        "\n",
        "import torch\n",
        "from dataclasses import dataclass\n",
        "from functools import partial\n",
        "import cv2\n",
        "import pandas as pd\n",
        "import gc\n",
        "import io\n",
        "import math\n",
        "import timm\n",
        "from IPython import display\n",
        "import lpips\n",
        "from PIL import Image, ImageOps\n",
        "import requests\n",
        "from glob import glob\n",
        "import json\n",
        "from types import SimpleNamespace\n",
        "from torch import nn\n",
        "from torch.nn import functional as F\n",
        "import torchvision.transforms as T\n",
        "import torchvision.transforms.functional as TF\n",
        "from tqdm.notebook import tqdm\n",
        "from CLIP import clip\n",
        "from resize_right import resize\n",
        "from guided_diffusion.script_util import create_model_and_diffusion, model_and_diffusion_defaults\n",
        "from datetime import datetime\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import random\n",
        "from ipywidgets import Output\n",
        "import hashlib\n",
        "from functools import partial\n",
        "if is_colab:\n",
        "    os.chdir('/content')\n",
        "    from google.colab import files\n",
        "else:\n",
        "    os.chdir(f'{PROJECT_DIR}')\n",
        "from IPython.display import Image as ipyimg\n",
        "from numpy import asarray\n",
        "from einops import rearrange, repeat\n",
        "import torch, torchvision\n",
        "import time\n",
        "from omegaconf import OmegaConf\n",
        "import warnings\n",
        "warnings.filterwarnings(\"ignore\", category=UserWarning)\n",
        "\n",
        "# AdaBins stuff\n",
        "if USE_ADABINS:\n",
        "    try:\n",
        "        from infer import InferenceHelper\n",
        "    except:\n",
        "        if not os.path.exists(\"AdaBins\"):\n",
        "            gitclone(\"https://github.com/shariqfarooq123/AdaBins.git\")\n",
        "        if not os.path.exists(f'{PROJECT_DIR}/pretrained/AdaBins_nyu.pt'):\n",
        "            createPath(f'{PROJECT_DIR}/pretrained')\n",
        "            wget(\"https://cloudflare-ipfs.com/ipfs/Qmd2mMnDLWePKmgfS8m6ntAg4nhV5VkUyAydYBp8cWWeB7/AdaBins_nyu.pt\", f'{PROJECT_DIR}/pretrained')\n",
        "        sys.path.append(f'{PROJECT_DIR}/AdaBins')\n",
        "    from infer import InferenceHelper\n",
        "    MAX_ADABINS_AREA = 500000\n",
        "\n",
        "import torch\n",
        "\n",
        "#@markdown use old pytorch broken for now :)\n",
        "use_old_pytorch = False #@param{type:'boolean'}\n",
        "if use_old_pytorch:\n",
        "  downgrade_pytorch_result = subprocess.run(['pip', 'install', 'torch==1.11.0', 'torchvision==0.12.0', 'torchaudio==0.11.0', '-q'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "  #downgrade_pytorch_result = subprocess.run(['conda', 'install', '-c', 'pytorch', 'pytorch=0.3.1', ])\n",
        "  downgrade_numpy_result = subprocess.run(['pip', 'uninstall', 'numpy'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "  downgrade_numpy_result = subprocess.run(['pip', 'install', 'numpy', 'numpy==1.20.0'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "\n",
        "\n",
        "# CUDA 11.3\n",
        "# conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch\n",
        "\n",
        "\n",
        "DEVICE = torch.device('cuda:0' if (torch.cuda.is_available() and not useCPU) else 'cpu')\n",
        "print('Using device:', DEVICE)\n",
        "device = DEVICE # At least one of the modules expects this name..\n",
        "\n",
        "\n",
        "if not useCPU:\n",
        "    if torch.cuda.get_device_capability(DEVICE) == (8,0): ## A100 fix thanks to Emad\n",
        "        print('Disabling CUDNN for A100 gpu', file=sys.stderr)\n",
        "        torch.backends.cudnn.enabled = False\n",
        "\n",
        "#zippy noise_injector stuff csa\n",
        "from PIL import Image\n",
        "import numpy as np\n",
        "from matplotlib import pyplot as plt\n",
        "import argparse\n",
        "import imutils\n",
        "import cv2\n",
        "from google.colab.patches import cv2_imshow\n",
        "from skimage import exposure"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "DefMidasFns"
      },
      "source": [
        "#@title ### 1.4 Define Midas functions\n",
        "\n",
        "from midas.dpt_depth import DPTDepthModel\n",
        "from midas.midas_net import MidasNet\n",
        "from midas.midas_net_custom import MidasNet_small\n",
        "from midas.transforms import Resize, NormalizeImage, PrepareForNet\n",
        "\n",
        "# Initialize MiDaS depth model.\n",
        "# It remains resident in VRAM and likely takes around 2GB VRAM.\n",
        "# You could instead initialize it for each frame (and free it after each frame) to save VRAM.. but initializing it is slow.\n",
        "default_models = {\n",
        "    \"midas_v21_small\": f\"{model_path}/midas_v21_small-70d6b9c8.pt\",\n",
        "    \"midas_v21\": f\"{model_path}/midas_v21-f6b98070.pt\",\n",
        "    \"dpt_large\": f\"{model_path}/dpt_large-midas-2f21e586.pt\",\n",
        "    \"dpt_hybrid\": f\"{model_path}/dpt_hybrid-midas-501f0c75.pt\",\n",
        "    \"dpt_hybrid_nyu\": f\"{model_path}/dpt_hybrid_nyu-2ce69ec7.pt\",}\n",
        "\n",
        "\n",
        "def init_midas_depth_model(midas_model_type=\"dpt_large\", optimize=True):\n",
        "    midas_model = None\n",
        "    net_w = None\n",
        "    net_h = None\n",
        "    resize_mode = None\n",
        "    normalization = None\n",
        "\n",
        "    print(f\"Initializing MiDaS '{midas_model_type}' depth model...\")\n",
        "    # load network\n",
        "    midas_model_path = default_models[midas_model_type]\n",
        "\n",
        "    if midas_model_type == \"dpt_large\": # DPT-Large\n",
        "        midas_model = DPTDepthModel(\n",
        "            path=midas_model_path,\n",
        "            backbone=\"vitl16_384\",\n",
        "            non_negative=True,\n",
        "        )\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode = \"minimal\"\n",
        "        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])\n",
        "    elif midas_model_type == \"dpt_hybrid\": #DPT-Hybrid\n",
        "        midas_model = DPTDepthModel(\n",
        "            path=midas_model_path,\n",
        "            backbone=\"vitb_rn50_384\",\n",
        "            non_negative=True,\n",
        "        )\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode=\"minimal\"\n",
        "        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])\n",
        "    elif midas_model_type == \"dpt_hybrid_nyu\": #DPT-Hybrid-NYU\n",
        "        midas_model = DPTDepthModel(\n",
        "            path=midas_model_path,\n",
        "            backbone=\"vitb_rn50_384\",\n",
        "            non_negative=True,\n",
        "        )\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode=\"minimal\"\n",
        "        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])\n",
        "    elif midas_model_type == \"midas_v21\":\n",
        "        midas_model = MidasNet(midas_model_path, non_negative=True)\n",
        "        net_w, net_h = 384, 384\n",
        "        resize_mode=\"upper_bound\"\n",
        "        normalization = NormalizeImage(\n",
        "            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]\n",
        "        )\n",
        "    elif midas_model_type == \"midas_v21_small\":\n",
        "        midas_model = MidasNet_small(midas_model_path, features=64, backbone=\"efficientnet_lite3\", exportable=True, non_negative=True, blocks={'expand': True})\n",
        "        net_w, net_h = 256, 256\n",
        "        resize_mode=\"upper_bound\"\n",
        "        normalization = NormalizeImage(\n",
        "            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]\n",
        "        )\n",
        "    else:\n",
        "        print(f\"midas_model_type '{midas_model_type}' not implemented\")\n",
        "        assert False\n",
        "\n",
        "    midas_transform = T.Compose(\n",
        "        [\n",
        "            Resize(\n",
        "                net_w,\n",
        "                net_h,\n",
        "                resize_target=None,\n",
        "                keep_aspect_ratio=True,\n",
        "                ensure_multiple_of=32,\n",
        "                resize_method=resize_mode,\n",
        "                image_interpolation_method=cv2.INTER_CUBIC,\n",
        "            ),\n",
        "            normalization,\n",
        "            PrepareForNet(),\n",
        "        ]\n",
        "    )\n",
        "\n",
        "    midas_model.eval()\n",
        "    \n",
        "    if optimize==True:\n",
        "        if DEVICE == torch.device(\"cuda\"):\n",
        "            midas_model = midas_model.to(memory_format=torch.channels_last)  \n",
        "            midas_model = midas_model.half()\n",
        "\n",
        "    midas_model.to(DEVICE)\n",
        "\n",
        "    print(f\"MiDaS '{midas_model_type}' depth model initialized.\")\n",
        "    return midas_model, midas_transform, net_w, net_h, resize_mode, normalization"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DefFns",
        "cellView": "form"
      },
      "source": [
        "#@title 1.5 Define necessary functions\n",
        "\n",
        "# https://gist.github.com/adefossez/0646dbe9ed4005480a2407c62aac8869\n",
        "\n",
        "import py3d_tools as p3dT\n",
        "import disco_xform_utils as dxf\n",
        "\n",
        "\n",
        "def interp(t):\n",
        "    return 3 * t**2 - 2 * t ** 3\n",
        "\n",
        "def perlin(width, height, scale=10, device=None):\n",
        "    gx, gy = torch.randn(2, width + 1, height + 1, 1, 1, device=device)\n",
        "    xs = torch.linspace(0, 1, scale + 1)[:-1, None].to(device)\n",
        "    ys = torch.linspace(0, 1, scale + 1)[None, :-1].to(device)\n",
        "    wx = 1 - interp(xs)\n",
        "    wy = 1 - interp(ys)\n",
        "    dots = 0\n",
        "    dots += wx * wy * (gx[:-1, :-1] * xs + gy[:-1, :-1] * ys)\n",
        "    dots += (1 - wx) * wy * (-gx[1:, :-1] * (1 - xs) + gy[1:, :-1] * ys)\n",
        "    dots += wx * (1 - wy) * (gx[:-1, 1:] * xs - gy[:-1, 1:] * (1 - ys))\n",
        "    dots += (1 - wx) * (1 - wy) * (-gx[1:, 1:] * (1 - xs) - gy[1:, 1:] * (1 - ys))\n",
        "    return dots.permute(0, 2, 1, 3).contiguous().view(width * scale, height * scale)\n",
        "\n",
        "def perlin_ms(octaves, width, height, grayscale, device=device):\n",
        "    out_array = [0.5] if grayscale else [0.5, 0.5, 0.5]\n",
        "    # out_array = [0.0] if grayscale else [0.0, 0.0, 0.0]\n",
        "    for i in range(1 if grayscale else 3):\n",
        "        scale = 2 ** len(octaves)\n",
        "        oct_width = width\n",
        "        oct_height = height\n",
        "        for oct in octaves:\n",
        "            p = perlin(oct_width, oct_height, scale, device)\n",
        "            out_array[i] += p * oct\n",
        "            scale //= 2\n",
        "            oct_width *= 2\n",
        "            oct_height *= 2\n",
        "    return torch.cat(out_array)\n",
        "\n",
        "def create_perlin_noise(octaves=[1, 1, 1, 1], width=2, height=2, grayscale=True):\n",
        "    out = perlin_ms(octaves, width, height, grayscale)\n",
        "    if grayscale:\n",
        "        out = TF.resize(size=(side_y, side_x), img=out.unsqueeze(0))\n",
        "        out = TF.to_pil_image(out.clamp(0, 1)).convert('RGB')\n",
        "    else:\n",
        "        out = out.reshape(-1, 3, out.shape[0]//3, out.shape[1])\n",
        "        out = TF.resize(size=(side_y, side_x), img=out)\n",
        "        out = TF.to_pil_image(out.clamp(0, 1).squeeze())\n",
        "\n",
        "    out = ImageOps.autocontrast(out)\n",
        "    return out\n",
        "\n",
        "def regen_perlin():\n",
        "    if perlin_mode == 'color':\n",
        "        init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "        init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, False)\n",
        "    elif perlin_mode == 'gray':\n",
        "        init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, True)\n",
        "        init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "    else:\n",
        "        init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "        init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "\n",
        "    init = TF.to_tensor(init).add(TF.to_tensor(init2)).div(2).to(device).unsqueeze(0).mul(2).sub(1)\n",
        "    del init2\n",
        "    return init.expand(batch_size, -1, -1, -1)\n",
        "\n",
        "def fetch(url_or_path):\n",
        "    if str(url_or_path).startswith('http://') or str(url_or_path).startswith('https://'):\n",
        "        r = requests.get(url_or_path)\n",
        "        r.raise_for_status()\n",
        "        fd = io.BytesIO()\n",
        "        fd.write(r.content)\n",
        "        fd.seek(0)\n",
        "        return fd\n",
        "    return open(url_or_path, 'rb')\n",
        "\n",
        "def read_image_workaround(path):\n",
        "    \"\"\"OpenCV reads images as BGR, Pillow saves them as RGB. Work around\n",
        "    this incompatibility to avoid colour inversions.\"\"\"\n",
        "    im_tmp = cv2.imread(path)\n",
        "    return cv2.cvtColor(im_tmp, cv2.COLOR_BGR2RGB)\n",
        "\n",
        "def parse_prompt(prompt):\n",
        "    if prompt.startswith('http://') or prompt.startswith('https://'):\n",
        "        vals = prompt.rsplit(':', 2)\n",
        "        vals = [vals[0] + ':' + vals[1], *vals[2:]]\n",
        "    else:\n",
        "        vals = prompt.rsplit(':', 1)\n",
        "    vals = vals + ['', '1'][len(vals):]\n",
        "    return vals[0], float(vals[1])\n",
        "\n",
        "\n",
        "def parse_prompt(prompt):\n",
        "    if prompt.startswith('http://') or prompt.startswith('https://'):\n",
        "        vals = prompt.rsplit(':', 2)\n",
        "        vals = [vals[0] + ':' + vals[1], *vals[2:]]\n",
        "    else:\n",
        "        vals = prompt.rsplit(':', 1)\n",
        "    vals = vals + ['', '1'][len(vals):]\n",
        "    vals[1] = eval(vals[1])\n",
        "    if isinstance(vals[1], int) or isinstance(vals[1], float):\n",
        "      return vals[0], [ float(x) for x in [float(vals[1])]*1000 ]\n",
        "    elif isinstance(vals[1], list):\n",
        "      return vals[0], [ float(x) for x in vals[1] ]\n",
        "    else:\n",
        "      raise TypeError(\"Only strings that evaluate to integers, floats and lists are allowed\")\n",
        "\n",
        "def sinc(x):\n",
        "    return torch.where(x != 0, torch.sin(math.pi * x) / (math.pi * x), x.new_ones([]))\n",
        "\n",
        "def lanczos(x, a):\n",
        "    cond = torch.logical_and(-a < x, x < a)\n",
        "    out = torch.where(cond, sinc(x) * sinc(x/a), x.new_zeros([]))\n",
        "    return out / out.sum()\n",
        "\n",
        "def ramp(ratio, width):\n",
        "    n = math.ceil(width / ratio + 1)\n",
        "    out = torch.empty([n])\n",
        "    cur = 0\n",
        "    for i in range(out.shape[0]):\n",
        "        out[i] = cur\n",
        "        cur += ratio\n",
        "    return torch.cat([-out[1:].flip([0]), out])[1:-1]\n",
        "\n",
        "def resample(input, size, align_corners=True):\n",
        "    n, c, h, w = input.shape\n",
        "    dh, dw = size\n",
        "\n",
        "    input = input.reshape([n * c, 1, h, w])\n",
        "\n",
        "    if dh < h:\n",
        "        kernel_h = lanczos(ramp(dh / h, 2), 2).to(input.device, input.dtype)\n",
        "        pad_h = (kernel_h.shape[0] - 1) // 2\n",
        "        input = F.pad(input, (0, 0, pad_h, pad_h), 'reflect')\n",
        "        input = F.conv2d(input, kernel_h[None, None, :, None])\n",
        "\n",
        "    if dw < w:\n",
        "        kernel_w = lanczos(ramp(dw / w, 2), 2).to(input.device, input.dtype)\n",
        "        pad_w = (kernel_w.shape[0] - 1) // 2\n",
        "        input = F.pad(input, (pad_w, pad_w, 0, 0), 'reflect')\n",
        "        input = F.conv2d(input, kernel_w[None, None, None, :])\n",
        "\n",
        "    input = input.reshape([n, c, h, w])\n",
        "    return F.interpolate(input, size, mode='bicubic', align_corners=align_corners)\n",
        "\n",
        "# class MakeCutouts(nn.Module):\n",
        "#     def __init__(self, cut_size, cutn, skip_augs=False):\n",
        "#         super().__init__()\n",
        "#         self.cut_size = cut_size\n",
        "#         self.cutn = cutn\n",
        "#         self.skip_augs = skip_augs\n",
        "#         self.augs = T.Compose([\n",
        "#             T.RandomHorizontalFlip(p=0.5),\n",
        "#             T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "#             T.RandomAffine(degrees=15, translate=(0.1, 0.1)),\n",
        "#             T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "#             T.RandomPerspective(distortion_scale=0.4, p=0.7),\n",
        "#             T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "#             T.RandomGrayscale(p=0.15),\n",
        "#             T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "#             # T.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1),\n",
        "#         ])\n",
        "\n",
        "#     def forward(self, input):\n",
        "#         input = T.Pad(input.shape[2]//4, fill=0)(input)\n",
        "#         sideY, sideX = input.shape[2:4]\n",
        "#         max_size = min(sideX, sideY)\n",
        "\n",
        "#         cutouts = []\n",
        "#         for ch in range(self.cutn):\n",
        "#             if ch > self.cutn - self.cutn//4:\n",
        "#                 cutout = input.clone()\n",
        "#             else:\n",
        "#                 size = int(max_size * torch.zeros(1,).normal_(mean=.8, std=.3).clip(float(self.cut_size/max_size), 1.))\n",
        "#                 offsetx = torch.randint(0, abs(sideX - size + 1), ())\n",
        "#                 offsety = torch.randint(0, abs(sideY - size + 1), ())\n",
        "#                 cutout = input[:, :, offsety:offsety + size, offsetx:offsetx + size]\n",
        "\n",
        "#             if not self.skip_augs:\n",
        "#                 cutout = self.augs(cutout)\n",
        "#             cutouts.append(resample(cutout, (self.cut_size, self.cut_size)))\n",
        "#             del cutout\n",
        "\n",
        "#         cutouts = torch.cat(cutouts, dim=0)\n",
        "#         return cutouts\n",
        "\n",
        "cutout_debug = False\n",
        "padargs = {}\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "max_size_cuts_replace_overview = False\n",
        "\n",
        "\n",
        "\n",
        "overview_weight = \"[1]*1000\"\n",
        "\n",
        "\n",
        "#random bucket pow or set??\n",
        "smallcut_size = 0\n",
        "smallcut_p = 0.35\n",
        "mediumcut_size = 0.08\n",
        "mediumcut_p = 0.5\n",
        "largecut_size = 0.2\n",
        "largecut_p = 0.7\n",
        "\n",
        "\n",
        "\n",
        "use_bucketpow = False\n",
        "use_rand_bucketpow = False\n",
        "\n",
        "ceiling = 1.0\n",
        "\n",
        "\n",
        "all_models_same_cuts = False\n",
        "\n",
        "\n",
        "use_floor = False\n",
        "\n",
        "adaptive_weight = False\n",
        "\n",
        "cut_weight_multiplier = \"\"\n",
        "\n",
        "manual_weight = False\n",
        "\n",
        "\n",
        "jaxy_overviews = False\n",
        "\n",
        "prompt_in_filename = False\n",
        "\n",
        "\n",
        "def randbucketpow():\n",
        "  rand = int(torch.rand([]))\n",
        "  if rand <= smallcut_p:\n",
        "    size_multiplier = smallcut_size\n",
        "  elif rand <= mediumcut_p and rand > smallcut_p:\n",
        "    size_multiplier = mediumcut_size\n",
        "  elif rand <= largecut_p and rand > mediumcut_p:\n",
        "    size_multiplier = largecut_size\n",
        "  elif rand > largecut_p:\n",
        "    size_multiplier = 2 \n",
        "  return size_multiplier\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "class MakeCutoutsES(nn.Module):\n",
        "    def __init__(self, cut_size, randoffsetx, randoffsety, randsize, randsize2, scheduler,\n",
        "                 Overview=4, InnerCrop = 0, IC_Size_Pow=0.5, IC_Grey = 0.2, overview_gray = 0.1, ov_proportion = 0.5, floor = 0.1, bucketcuts = 16\n",
        "                 ):\n",
        "        super().__init__()\n",
        "        self.cut_size = cut_size\n",
        "        self.Overview = Overview\n",
        "        self.InnerCrop = InnerCrop\n",
        "        self.IC_Size_Pow = IC_Size_Pow\n",
        "        self.IC_Grey = IC_Grey\n",
        "        self.overview_gray = overview_gray\n",
        "        self.ov_proportion = ov_proportion\n",
        "        self.floor = floor\n",
        "        self.randoffsetx = randoffsetx\n",
        "        self.randoffsety = randoffsety\n",
        "        self.randsize = randsize\n",
        "        self.randsize2 = randsize2\n",
        "        self.bucketcuts = bucketcuts\n",
        "        self.sizes = []\n",
        "\n",
        "\n",
        "        if args.animation_mode == 'None':\n",
        "          self.ic_augs = T.Compose([\n",
        "                T.RandomHorizontalFlip(p=0.5),\n",
        "                #T.Lambda(lambda x: x - (pixel_jitter/2) + torch.randn_like(x) * pixel_jitter),\n",
        "                T.Lambda(lambda x: x + torch.randn_like(x) * pixel_jitter),\n",
        "                T.ColorJitter(brightness=col_jitter, contrast=col_jitter, saturation=col_jitter, hue=col_jitter),\n",
        "              ])\n",
        "          tmpaugs = [\n",
        "              #T.Lambda(lambda x: x - (pixel_jitter/2) + torch.randn_like(x) * pixel_jitter),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * pixel_jitter),\n",
        "              T.ColorJitter(brightness=col_jitter, contrast=col_jitter, saturation=col_jitter, hue=col_jitter),\n",
        "          ]\n",
        "          if (hflip ==True):\n",
        "            tmpaugs.append(T.RandomHorizontalFlip(p=0.5))\n",
        "          if (perspective == True):\n",
        "            tmpaugs.append(T.RandomPerspective(distortion_scale=0.18, p=0.35))\n",
        "          if affine == True and scheduler > (1000 - percent_stop_affine*10):\n",
        "            tmpaugs.append(T.RandomAffine(degrees=degree_tilt, translate=(0.05, 0.05),  interpolation = T.InterpolationMode.BILINEAR))\n",
        "          \n",
        "          self.ov_augs = T.Compose(tmpaugs)\n",
        "\n",
        "        elif args.animation_mode == 'Video Input':\n",
        "          self.ov_augs = T.Compose([\n",
        "              T.RandomHorizontalFlip(p=0.5),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomAffine(degrees=15, translate=(0.1, 0.1)),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomPerspective(distortion_scale=0.4, p=0.7),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomGrayscale(p=0.15),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              # T.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.1),\n",
        "          ])\n",
        "          self.ic_augs = self.ov_augs\n",
        "        elif  args.animation_mode == '2D' or args.animation_mode == '3D':\n",
        "          self.ov_augs = T.Compose([\n",
        "              T.RandomHorizontalFlip(p=0.4),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomAffine(degrees=10, translate=(0.05, 0.05),  interpolation = T.InterpolationMode.BILINEAR),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.RandomGrayscale(p=0.1),\n",
        "              T.Lambda(lambda x: x + torch.randn_like(x) * 0.01),\n",
        "              T.ColorJitter(brightness=0.1, contrast=0.1, saturation=0.1, hue=0.3),\n",
        "          ])\n",
        "          self.ic_augs = self.ov_augs\n",
        "          \n",
        "\n",
        "    def forward(self, input):\n",
        "        cutouts = []\n",
        "        counter = 0\n",
        "        gray = T.Grayscale(3)\n",
        "        sideY, sideX = input.shape[2:4]\n",
        "        max_size = min(sideX, sideY)\n",
        "        min_size = min(sideX, sideY, self.cut_size)\n",
        "        floor_size = min_size + (max_size - min_size) * self.floor\n",
        "        l_size = max(sideX, sideY)\n",
        "        output_shape = [1,3,self.cut_size,self.cut_size] \n",
        "        output_shape_2 = [1,3,self.cut_size+2,self.cut_size+2]\n",
        "        pad_input = F.pad(input,((sideY-max_size)//2,(sideY-max_size)//2,(sideX-max_size)//2,(sideX-max_size)//2), **padargs)\n",
        "        cutout = resize(pad_input, out_shape=output_shape)\n",
        "        bordery = ((sideY - max_size)//2)*2\n",
        "        borderx = ((sideX - max_size)//2)*2\n",
        "        border = int(((sideX*sideY)**0.5)*0.1)\n",
        "\n",
        "        if use_bucketpow:\n",
        "          for i in range(self.bucketcuts):\n",
        "            if (i+1)/bucketcuts >= largecut_p:\n",
        "              cutout = resize(pad_input, out_shape=output_shape)\n",
        "              if (torch.rand([]) < self.overview_gray):\n",
        "                cutout = gray(cutout)\n",
        "            else:\n",
        "              if (i+1)/bucketcuts <= smallcut_p:\n",
        "                size = int(smallcut_size*(max_size - min_size) + min_size)\n",
        "              elif (i+1)/bucketcuts <= mediumcut_p and (i+1)/bucketcuts > smallcut_p:\n",
        "                size = int(mediumcut_size*(max_size - min_size) + min_size)\n",
        "              elif (i+1)/bucketcuts < largecut_p and (i+1)/bucketcuts > mediumcut_p:\n",
        "                size = int(largecut_size*(max_size - min_size) + min_size) \n",
        "              offsetx = int((self.randoffsetx[counter])*(sideX - size + 1))\n",
        "              offsety = int((self.randoffsety[counter])*(sideY - size + 1))\n",
        "              cutout = input[:, :, offsety:offsety + size, offsetx:offsetx + size]\n",
        "              if (torch.rand([]) < self.IC_Grey):\n",
        "                cutout = gray(cutout)\n",
        "              cutout = resize(cutout, out_shape=output_shape)\n",
        "            if not skip_augs:\n",
        "              cutouts.append(self.ic_augs(cutout))\n",
        "            else: \n",
        "              cutouts.append(cutout)\n",
        "            counter += 1\n",
        "            if adaptive_weight:\n",
        "              self.sizes.append(max(sideX,sideY)**2)\n",
        "            \n",
        "        else:\n",
        "          if self.Overview > 0:\n",
        "            #cutout = resize(pad_input, out_shape=output_shape)\n",
        "            for j in range(int(math.ceil(self.Overview*self.ov_proportion))):\n",
        "                randborderx = int(self.randoffsetx[counter]*border)\n",
        "                randbordery = int(self.randoffsety[counter]*border)\n",
        "                if jaxy_overviews:\n",
        "                  my_pad = F.pad(input,(math.ceil(bordery * self.randsize2[counter]) + 20 + math.ceil(randbordery * self.randsize[counter]),math.floor(bordery * (1 - self.randsize2[counter])) + 20 + math.floor((border - randbordery)*self.randsize[counter]),math.ceil(borderx * self.randsize2[counter]) + 20 + math.ceil(randborderx * self.randsize[counter]),math.floor(borderx * (1-self.randsize2[counter])) + 20 + math.floor((border - randborderx)*self.randsize[counter])), **padargs)\n",
        "                  cutout = resize(my_pad, out_shape=output_shape)\n",
        "                elif jaxy_overviews2:\n",
        "                  my_pad = F.pad(input,((sideY-max_size)//2 + 20 + math.ceil(randbordery * self.randsize[counter]),(sideY-max_size)//2 + 20 + math.floor((80 - randbordery)*self.randsize[counter]),(sideX-max_size)//2 + 20 + math.ceil(randborderx * self.randsize[counter]),(sideX-max_size)//2 + 20 + math.floor((80 - randborderx)*self.randsize[counter])), **padargs)\n",
        "                  cutout = resize(my_pad, out_shape=output_shape)\n",
        "                elif zoom_out_overviews:\n",
        "                  my_pad = F.pad(input,((sideY-max_size)//2 + 10 + math.ceil(border * self.randsize[counter]),(sideY-max_size)//2 + 10 + math.floor(border*self.randsize[counter]),(sideX-max_size)//2 + 10 + math.ceil(border * self.randsize[counter]),(sideX-max_size)//2 + 10 + math.floor(border*self.randsize[counter])), **padargs) \n",
        "                  cutout = resize(my_pad, out_shape=output_shape)\n",
        "                else:\n",
        "                  cutout = resize(pad_input, out_shape=output_shape)\n",
        "                if (torch.rand([]) < self.overview_gray):\n",
        "                    cutout = gray(cutout)\n",
        "                if not skip_augs:\n",
        "                    cutouts.append(self.ov_augs(cutout))\n",
        "                else: \n",
        "                    cutouts.append(cutout)\n",
        "                counter += 1\n",
        "                if adaptive_weight:\n",
        "                  self.sizes.append(max(sideX,sideY)**2)\n",
        "            for k in range(int(math.floor(self.Overview*(1-self.ov_proportion)))):\n",
        "              if (self.randsize[counter] < 0.8):\n",
        "                size = max_size\n",
        "              else:\n",
        "                size = int((max_size - min_size)*0.7 + min_size)\n",
        "              offsetx = int((self.randoffsetx[counter])*(sideX - size + 1))\n",
        "              offsety = int((self.randoffsety[counter])*(sideY - size + 1))\n",
        "              cutout = input[:, :, offsety:offsety + size, offsetx:offsetx + size]\n",
        "              if (torch.rand([]) < self.overview_gray):\n",
        "                  cutout = gray(cutout)\n",
        "              cutout = resize(cutout, out_shape=output_shape)\n",
        "              if not skip_augs:\n",
        "                  cutouts.append(self.ic_augs(cutout))\n",
        "              else: \n",
        "                  cutouts.append(cutout)\n",
        "              counter += 1\n",
        "              if adaptive_weight:\n",
        "                self.sizes.append(size**2)\n",
        "\n",
        "            if cutout_debug:\n",
        "                if is_colab:\n",
        "                    TF.to_pil_image(cutouts[0].clamp(0, 1).squeeze(0)).save(\"/content/cutout_overview0.jpg\",quality=99)\n",
        "                else:\n",
        "                    TF.to_pil_image(cutouts[0].clamp(0, 1).squeeze(0)).save(\"cutout_overview0.jpg\",quality=99)\n",
        "\n",
        "                                \n",
        "          if self.InnerCrop >0:\n",
        "              for i in range(self.InnerCrop):\n",
        "                  if use_floor:\n",
        "                    size = int(self.randsize[counter]**self.IC_Size_Pow * ceiling * (max_size - floor_size) + floor_size)\n",
        "                  else:\n",
        "                    size = int(self.randsize[counter]**self.IC_Size_Pow * ceiling * (max_size - min_size) + min_size)  \n",
        "                  offsetx = int((self.randoffsetx[counter])*(sideX - size + 1))\n",
        "                  offsety = int((self.randoffsety[counter])*(sideY - size + 1))\n",
        "                  cutout = input[:, :, offsety:offsety + size, offsetx:offsetx + size]\n",
        "                  if (torch.rand([]) < self.IC_Grey):\n",
        "                      cutout = gray(cutout)\n",
        "                  cutout = resize(cutout, out_shape=output_shape)\n",
        "                  if not skip_augs:\n",
        "                      cutouts.append(self.ic_augs(cutout))\n",
        "                  else: \n",
        "                      cutouts.append(cutout)\n",
        "                  counter += 1\n",
        "                  if adaptive_weight:\n",
        "                    self.sizes.append(size**2)\n",
        "\n",
        "              if cutout_debug:\n",
        "                  if is_colab:\n",
        "                      TF.to_pil_image(cutouts[-1].clamp(0, 1).squeeze(0)).save(\"/content/cutout_InnerCrop.jpg\",quality=99)\n",
        "                  else:\n",
        "                      TF.to_pil_image(cutouts[-1].clamp(0, 1).squeeze(0)).save(\"cutout_InnerCrop.jpg\",quality=99)\n",
        "        cutouts = torch.cat(cutouts)\n",
        "        return cutouts\n",
        "\n",
        "def spherical_dist_loss(x, y):\n",
        "    x = F.normalize(x, dim=-1)\n",
        "    y = F.normalize(y, dim=-1)\n",
        "    return (x - y).norm(dim=-1).div(2).arcsin().pow(2).mul(2)     \n",
        "\n",
        "def symm_loss(im,lpm):\n",
        "    h = int(im.shape[3]/2)\n",
        "    h1,h2 = im[:,:,:,:h],im[:,:,:,h:]\n",
        "    h2 = TF.hflip(h2)\n",
        "    return lpm(h1,h2) \n",
        "\n",
        "def tv_loss(input):\n",
        "    \"\"\"L2 total variation loss, as in Mahendran et al.\"\"\"\n",
        "    input = F.pad(input, (0, 1, 0, 1), 'replicate')\n",
        "    x_diff = input[..., :-1, 1:] - input[..., :-1, :-1]\n",
        "    y_diff = input[..., 1:, :-1] - input[..., :-1, :-1]\n",
        "    return (x_diff**2 + y_diff**2).mean([1, 2, 3])\n",
        "\n",
        "\n",
        "def range_loss(input):\n",
        "    return (input - input.clamp(-1, 1)).pow(2).mean([1, 2, 3])\n",
        "\n",
        "stop_on_next_loop = False  # Make sure GPU memory doesn't get corrupted from cancelling the run mid-way through, allow a full frame to complete\n",
        "TRANSLATION_SCALE = 1.0/200.0\n",
        "\n",
        "def do_3d_step(img_filepath, frame_num, midas_model, midas_transform):\n",
        "  if args.key_frames:\n",
        "    translation_x = args.translation_x_series[frame_num]\n",
        "    translation_y = args.translation_y_series[frame_num]\n",
        "    translation_z = args.translation_z_series[frame_num]\n",
        "    rotation_3d_x = args.rotation_3d_x_series[frame_num]\n",
        "    rotation_3d_y = args.rotation_3d_y_series[frame_num]\n",
        "    rotation_3d_z = args.rotation_3d_z_series[frame_num]\n",
        "    print(\n",
        "        f'translation_x: {translation_x}',\n",
        "        f'translation_y: {translation_y}',\n",
        "        f'translation_z: {translation_z}',\n",
        "        f'rotation_3d_x: {rotation_3d_x}',\n",
        "        f'rotation_3d_y: {rotation_3d_y}',\n",
        "        f'rotation_3d_z: {rotation_3d_z}',\n",
        "    )\n",
        "\n",
        "  translate_xyz = [-translation_x*TRANSLATION_SCALE, translation_y*TRANSLATION_SCALE, -translation_z*TRANSLATION_SCALE]\n",
        "  rotate_xyz_degrees = [rotation_3d_x, rotation_3d_y, rotation_3d_z]\n",
        "  print('translation:',translate_xyz)\n",
        "  print('rotation:',rotate_xyz_degrees)\n",
        "  rotate_xyz = [math.radians(rotate_xyz_degrees[0]), math.radians(rotate_xyz_degrees[1]), math.radians(rotate_xyz_degrees[2])]\n",
        "  rot_mat = p3dT.euler_angles_to_matrix(torch.tensor(rotate_xyz, device=device), \"XYZ\").unsqueeze(0)\n",
        "  print(\"rot_mat: \" + str(rot_mat))\n",
        "  next_step_pil = dxf.transform_image_3d(img_filepath, midas_model, midas_transform, DEVICE,\n",
        "                                          rot_mat, translate_xyz, args.near_plane, args.far_plane,\n",
        "                                          args.fov, padding_mode=args.padding_mode,\n",
        "                                          sampling_mode=args.sampling_mode, midas_weight=args.midas_weight)\n",
        "  return next_step_pil\n",
        "\n",
        "def symmetry_transformation_fn(x):\n",
        "    if args.use_horizontal_symmetry:\n",
        "        [n, c, h, w] = x.size()\n",
        "        x = torch.concat((x[:, :, :, :w//2], torch.flip(x[:, :, :, :w//2], [-1])), -1)\n",
        "        print(\"horizontal symmetry applied\")\n",
        "    if args.use_vertical_symmetry:\n",
        "        [n, c, h, w] = x.size()\n",
        "        x = torch.concat((x[:, :, :h//2, :], torch.flip(x[:, :, :h//2, :], [-2])), -2)\n",
        "        print(\"vertical symmetry applied\")\n",
        "    return x\n",
        "\n",
        "def resize_transformation_fn(image):\n",
        "  if args.resize_mid_run:\n",
        "    image = resize(image, out_shape=[1,3,newy,newx])\n",
        "    side_x = newx\n",
        "    side_y = newy\n",
        "  return image\n",
        "\n",
        "def do_run():\n",
        "  seed = args.seed\n",
        "  print(range(args.start_frame, args.max_frames))\n",
        "\n",
        "  if (args.animation_mode == \"3D\") and (args.midas_weight > 0.0):\n",
        "      midas_model, midas_transform, midas_net_w, midas_net_h, midas_resize_mode, midas_normalization = init_midas_depth_model(args.midas_depth_model)\n",
        "  for frame_num in range(args.start_frame, args.max_frames):\n",
        "      if stop_on_next_loop:\n",
        "        break\n",
        "      \n",
        "      display.clear_output(wait=True)\n",
        "\n",
        "      # Print Frame progress if animation mode is on\n",
        "      if args.animation_mode != \"None\":\n",
        "        batchBar = tqdm(range(args.max_frames), desc =\"Frames\")\n",
        "        batchBar.n = frame_num\n",
        "        batchBar.refresh()\n",
        "\n",
        "      \n",
        "      # Inits if not video frames\n",
        "      if args.animation_mode != \"Video Input\":\n",
        "        if args.init_image in ['','none', 'None', 'NONE']:\n",
        "          init_image = None\n",
        "        else:\n",
        "          init_image = args.init_image\n",
        "        init_scale = args.init_scale\n",
        "        skip_steps = args.skip_steps\n",
        "\n",
        "      if args.animation_mode == \"2D\":\n",
        "        if args.key_frames:\n",
        "          angle = args.angle_series[frame_num]\n",
        "          zoom = args.zoom_series[frame_num]\n",
        "          translation_x = args.translation_x_series[frame_num]\n",
        "          translation_y = args.translation_y_series[frame_num]\n",
        "          print(\n",
        "              f'angle: {angle}',\n",
        "              f'zoom: {zoom}',\n",
        "              f'translation_x: {translation_x}',\n",
        "              f'translation_y: {translation_y}',\n",
        "          )\n",
        "        \n",
        "        if frame_num > 0:\n",
        "          seed += 1\n",
        "          if resume_run and frame_num == start_frame:\n",
        "            img_0 = cv2.imread(batchFolder+f\"/{batch_name}({batchNum})_{start_frame-1:04}.png\")\n",
        "          else:\n",
        "            img_0 = cv2.imread('prevFrame.png')\n",
        "          center = (1*img_0.shape[1]//2, 1*img_0.shape[0]//2)\n",
        "          trans_mat = np.float32(\n",
        "              [[1, 0, translation_x],\n",
        "              [0, 1, translation_y]]\n",
        "          )\n",
        "          rot_mat = cv2.getRotationMatrix2D( center, angle, zoom )\n",
        "          trans_mat = np.vstack([trans_mat, [0,0,1]])\n",
        "          rot_mat = np.vstack([rot_mat, [0,0,1]])\n",
        "          transformation_matrix = np.matmul(rot_mat, trans_mat)\n",
        "          img_0 = cv2.warpPerspective(\n",
        "              img_0,\n",
        "              transformation_matrix,\n",
        "              (img_0.shape[1], img_0.shape[0]),\n",
        "              borderMode=cv2.BORDER_WRAP\n",
        "          )\n",
        "\n",
        "          cv2.imwrite('prevFrameScaled.png', img_0)\n",
        "          init_image = 'prevFrameScaled.png'\n",
        "          init_scale = args.frames_scale\n",
        "          skip_steps = args.calc_frames_skip_steps\n",
        "\n",
        "      if args.animation_mode == \"3D\":\n",
        "        if frame_num > 0:\n",
        "          seed += 1    \n",
        "          if resume_run and frame_num == start_frame:\n",
        "            img_filepath = batchFolder+f\"/{batch_name}({batchNum})_{start_frame-1:04}.png\"\n",
        "            if turbo_mode and frame_num > turbo_preroll:\n",
        "              shutil.copyfile(img_filepath, 'oldFrameScaled.png')\n",
        "          else:\n",
        "            img_filepath = '/content/prevFrame.png' if is_colab else 'prevFrame.png'\n",
        "\n",
        "          next_step_pil = do_3d_step(img_filepath, frame_num, midas_model, midas_transform)\n",
        "          next_step_pil.save('prevFrameScaled.png')\n",
        "\n",
        "          ### Turbo mode - skip some diffusions, use 3d morph for clarity and to save time\n",
        "          if turbo_mode:\n",
        "            if frame_num == turbo_preroll: #start tracking oldframe\n",
        "              next_step_pil.save('oldFrameScaled.png')#stash for later blending          \n",
        "            elif frame_num > turbo_preroll:\n",
        "              #set up 2 warped image sequences, old & new, to blend toward new diff image\n",
        "              old_frame = do_3d_step('oldFrameScaled.png', frame_num, midas_model, midas_transform)\n",
        "              old_frame.save('oldFrameScaled.png')\n",
        "              if frame_num % int(turbo_steps) != 0: \n",
        "                print('turbo skip this frame: skipping clip diffusion steps')\n",
        "                filename = f'{args.batch_name}({args.batchNum})_{frame_num:04}.png'\n",
        "                blend_factor = ((frame_num % int(turbo_steps))+1)/int(turbo_steps)\n",
        "                print('turbo skip this frame: skipping clip diffusion steps and saving blended frame')\n",
        "                newWarpedImg = cv2.imread('prevFrameScaled.png')#this is already updated..\n",
        "                oldWarpedImg = cv2.imread('oldFrameScaled.png')\n",
        "                blendedImage = cv2.addWeighted(newWarpedImg, blend_factor, oldWarpedImg,1-blend_factor, 0.0)\n",
        "                cv2.imwrite(f'{batchFolder}/{filename}',blendedImage)\n",
        "                next_step_pil.save(f'{img_filepath}') # save it also as prev_frame to feed next iteration\n",
        "                if vr_mode:\n",
        "                  generate_eye_views(TRANSLATION_SCALE,batchFolder,filename,frame_num,midas_model, midas_transform)\n",
        "                continue\n",
        "              else:\n",
        "                #if not a skip frame, will run diffusion and need to blend.\n",
        "                oldWarpedImg = cv2.imread('prevFrameScaled.png')\n",
        "                cv2.imwrite(f'oldFrameScaled.png',oldWarpedImg)#swap in for blending later \n",
        "                print('clip/diff this frame - generate clip diff image')\n",
        "\n",
        "          init_image = 'prevFrameScaled.png'\n",
        "          init_scale = args.frames_scale\n",
        "          skip_steps = args.calc_frames_skip_steps\n",
        "\n",
        "      if  args.animation_mode == \"Video Input\":\n",
        "        init_scale = args.video_init_frames_scale\n",
        "        skip_steps = args.calc_frames_skip_steps\n",
        "        if not video_init_seed_continuity:\n",
        "          seed += 1\n",
        "        if video_init_flow_warp:\n",
        "          if frame_num == 0: \n",
        "            skip_steps = args.video_init_skip_steps\n",
        "            init_image = f'{videoFramesFolder}/{frame_num+1:04}.jpg'\n",
        "          if frame_num > 0: \n",
        "            prev = PIL.Image.open(batchFolder+f\"/{batch_name}({batchNum})_{frame_num-1:04}.png\")\n",
        "            \n",
        "            frame1_path = f'{videoFramesFolder}/{frame_num:04}.jpg'\n",
        "            frame2 = PIL.Image.open(f'{videoFramesFolder}/{frame_num+1:04}.jpg')\n",
        "            flo_path = f\"/{flo_folder}/{frame1_path.split('/')[-1]}.npy\"\n",
        "            \n",
        "            init_image = 'warped.png'\n",
        "            print(video_init_flow_blend)\n",
        "            weights_path = None\n",
        "            if video_init_check_consistency:\n",
        "                # TBD\n",
        "                pass\n",
        " \n",
        "            warp(prev, frame2, flo_path, blend=video_init_flow_blend, weights_path=weights_path).save(init_image)\n",
        "            \n",
        "        else:\n",
        "          init_image = f'{videoFramesFolder}/{frame_num+1:04}.jpg'\n",
        "\n",
        "\n",
        "      loss_values = []\n",
        "  \n",
        "      if seed is not None:\n",
        "          np.random.seed(seed)\n",
        "          random.seed(seed)\n",
        "          torch.manual_seed(seed)\n",
        "          torch.cuda.manual_seed_all(seed)\n",
        "          torch.backends.cudnn.deterministic = True\n",
        "  \n",
        "      target_embeds, weights = [], []\n",
        "      \n",
        "\n",
        "      if args.prompts_series is not None and frame_num >= len(args.prompts_series):\n",
        "        frame_prompt = args.prompts_series[-1]\n",
        "      elif args.prompts_series is not None:\n",
        "        frame_prompt = args.prompts_series[frame_num]\n",
        "      else:\n",
        "        frame_prompt = []\n",
        "      \n",
        "      print(args.image_prompts_series)\n",
        "      if args.image_prompts_series is not None and frame_num >= len(args.image_prompts_series):\n",
        "        image_prompt = args.image_prompts_series[-1]\n",
        "      elif args.image_prompts_series is not None:\n",
        "        image_prompt = args.image_prompts_series[frame_num]\n",
        "      else:\n",
        "        image_prompt = []\n",
        "\n",
        "      print(f'Frame {frame_num} Prompt: {frame_prompt}')\n",
        "\n",
        "      model_stats = []\n",
        "      temp_weight = []\n",
        "      for clip_model in clip_models:\n",
        "            model_stat = {\"clip_model\":None,\"target_embeds\":[],\"make_cutouts\":None,\"weights\":[]}\n",
        "            model_stat[\"clip_model\"] = clip_model\n",
        "            \n",
        "            for prompt in frame_prompt:\n",
        "                txt, weight = parse_prompt(prompt)\n",
        "                txt = clip_model.encode_text(clip.tokenize(prompt).to(device)).float()\n",
        "                \n",
        "                if args.fuzzy_prompt:\n",
        "                    for i in range(25):\n",
        "                        model_stat[\"target_embeds\"].append((txt + torch.randn(txt.shape).cuda() * args.rand_mag).clamp(0,1))\n",
        "                        model_stat[\"weights\"].append(weight)\n",
        "                else:\n",
        "                    model_stat[\"target_embeds\"].append(txt)\n",
        "                    model_stat[\"weights\"].append(weight)\n",
        "        \n",
        "            if image_prompt:\n",
        "              randomx = torch.rand(16)\n",
        "              randomy = torch.rand(16)\n",
        "              randomsize = torch.rand(16)\n",
        "              randomsize2 = torch.rand(16)\n",
        "              model_stat[\"make_cutouts\"] = MakeCutoutsES(clip_model.visual.input_resolution,\n",
        "                            randoffsetx = randomx,\n",
        "                            randoffsety = randomy,\n",
        "                            randsize = randomsize,\n",
        "                            randsize2 = randomsize2,\n",
        "                            scheduler = 1000,\n",
        "                            Overview = 8, \n",
        "                            InnerCrop = 8,\n",
        "                            bucketcuts = 0,\n",
        "                            IC_Size_Pow=0.24,\n",
        "                            IC_Grey = 0.0,\n",
        "                            overview_gray = 0.0,\n",
        "                            ov_proportion = 1.0,\n",
        "                            floor = 0.0) \n",
        "              for prompt in image_prompt:\n",
        "                  path, weight = parse_prompt(prompt)\n",
        "                  img = Image.open(fetch(path)).convert('RGB')\n",
        "                  img = TF.resize(img, min(side_x, side_y, *img.size), T.InterpolationMode.LANCZOS)\n",
        "                  batch = model_stat[\"make_cutouts\"](TF.to_tensor(img).to(device).unsqueeze(0).mul(2).sub(1))\n",
        "                  #passes = 4\n",
        "                  #total = model_stat[\"make_cutouts\"](TF.to_tensor(img).to(device).unsqueeze(0).mul(2).sub(1))\n",
        "                  #for i in range(passes):\n",
        "                    #batch = model_stat[\"make_cutouts\"](TF.to_tensor(img).to(device).unsqueeze(0).mul(2).sub(1))\n",
        "                    #total = total + batch\n",
        "                  #embed = clip_model.encode_image(normalize(total)).float()\n",
        "                  embed = clip_model.encode_image(normalize(batch)).float()\n",
        "                  if fuzzy_prompt:\n",
        "                      for i in range(25):\n",
        "                          model_stat[\"target_embeds\"].append((embed + torch.randn(embed.shape).cuda() * rand_mag).clamp(0,1))\n",
        "                          weights.extend([weight / 16] * 16)\n",
        "                  else:\n",
        "                      model_stat[\"target_embeds\"].append(embed)\n",
        "                      model_stat[\"weights\"].extend([weight / 16] * 16)\n",
        "        \n",
        "            model_stat[\"target_embeds\"] = torch.cat(model_stat[\"target_embeds\"])\n",
        "            # model_stat[\"weights\"] = torch.tensor(model_stat[\"weights\"], device=device)\n",
        "            # if model_stat[\"weights\"].sum().abs() < 1e-3:\n",
        "            #     raise RuntimeError('The weights must not sum to 0.')\n",
        "            # model_stat[\"weights\"] /= model_stat[\"weights\"].sum().abs()\n",
        "            model_stats.append(model_stat)\n",
        "  \n",
        "      init = None\n",
        "      if init_image is not None:\n",
        "          init = Image.open(fetch(init_image)).convert('RGB')\n",
        "          #zippy noise injector - only for animation frames csa\n",
        "          if (args.noise_injector == True or args.copy_palette == True) and frame_num > 0:\n",
        "            init_to_noise = cv2.imread(init_image)\n",
        "            if args.copy_palette == True:\n",
        "              palette_source_img = cv2.imread(args.copy_palette_image)\n",
        "              init_to_noise = exposure.match_histograms(init_to_noise, palette_source_img, multichannel=True) \n",
        "            if args.noise_injector == True:\n",
        "              edge_mix = get_edge_mix(center_crop(init_to_noise,.33))\n",
        "              print('edge mix is...',round(edge_mix,1))\n",
        "              if edge_mix < args.noise_injector_threshold:\n",
        "                init_to_noise = add_noise(init_to_noise,args.noise_injector_mix,args.noise_injector_px_size,\n",
        "                                          args.noise_injector_spice,args.noise_injector_mask_feather,\n",
        "                                          args.noise_injector_sigma) \n",
        "                print('edge mix was below threshold, noise was added')\n",
        "                #ratio = np.amax(init_to_noise) / 256       \n",
        "                #converted_noised_init = (init_to_noise / ratio).astype('uint8')\n",
        "                #recolored_noised_init = cv2.cvtColor(converted_noised_init,cv2.COLOR_BGR2RGB)\n",
        "                #init = Image.fromarray(recolored_noised_init)\n",
        "              else:\n",
        "                print('edge mix was above threshold, no noise was added')\n",
        "            #now convert init_to_noise back to PIL img format\n",
        "            ratio = np.amax(init_to_noise) / 256       \n",
        "            converted_noised_init = (init_to_noise / ratio).astype('uint8')\n",
        "            recolored_noised_init = cv2.cvtColor(converted_noised_init,cv2.COLOR_BGR2RGB)\n",
        "            init = Image.fromarray(recolored_noised_init)         \n",
        "          #end noise injector\n",
        "          init = init.resize((args.side_x, args.side_y), Image.LANCZOS)\n",
        "          init = TF.to_tensor(init).to(device).unsqueeze(0).mul(2).sub(1)\n",
        "      \n",
        "      if args.perlin_init:\n",
        "          if args.perlin_mode == 'color':\n",
        "              init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "              init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, False)\n",
        "          elif args.perlin_mode == 'gray':\n",
        "            init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, True)\n",
        "            init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "          else:\n",
        "            init = create_perlin_noise([1.5**-i*0.5 for i in range(12)], 1, 1, False)\n",
        "            init2 = create_perlin_noise([1.5**-i*0.5 for i in range(8)], 4, 4, True)\n",
        "          # init = TF.to_tensor(init).add(TF.to_tensor(init2)).div(2).to(device)\n",
        "          init = TF.to_tensor(init).add(TF.to_tensor(init2)).div(2).to(device).unsqueeze(0).mul(2).sub(1)\n",
        "          del init2\n",
        "  \n",
        "      cur_t = None\n",
        "  \n",
        "      def cond_fn(x, t, y=None):\n",
        "          with torch.enable_grad():\n",
        "              x_is_NaN = False\n",
        "              x = x.detach().requires_grad_()\n",
        "              n = x.shape[0]\n",
        "              t_int = int(t.item()) + 1\n",
        "              if args.use_secondary_model[1000-t_int] is True:\n",
        "                alpha = torch.tensor(diffusion.sqrt_alphas_cumprod[cur_t], device=device, dtype=torch.float32)\n",
        "                sigma = torch.tensor(diffusion.sqrt_one_minus_alphas_cumprod[cur_t], device=device, dtype=torch.float32)\n",
        "                cosine_t = alpha_sigma_to_t(alpha, sigma)\n",
        "                out = secondary_model(x, cosine_t[None].repeat([n])).pred\n",
        "                fac = diffusion.sqrt_one_minus_alphas_cumprod[cur_t]\n",
        "                x_in = out * fac + x * (1 - fac)\n",
        "                x_in_grad = torch.zeros_like(x_in)\n",
        "              else:\n",
        "                my_t = torch.ones([n], device=device, dtype=torch.long) * cur_t\n",
        "                out = diffusion.p_mean_variance(model, x, my_t, clip_denoised=False, model_kwargs={'y': y})\n",
        "                fac = diffusion.sqrt_one_minus_alphas_cumprod[cur_t]\n",
        "                x_in = out['pred_xstart'] * fac + x * (1 - fac)\n",
        "                x_in_grad = torch.zeros_like(x_in)\n",
        "              for i in range(args.cutn_batches[1000 - t_int]):\n",
        "                randomx = torch.rand((args.cut_overview[1000-t_int]+args.cut_innercut[1000-t_int]))\n",
        "                randomy = torch.rand((args.cut_overview[1000-t_int]+args.cut_innercut[1000-t_int]))\n",
        "                randomsize = torch.rand((args.cut_overview[1000-t_int]+args.cut_innercut[1000-t_int]))\n",
        "                randomsize2 = torch.rand((args.cut_overview[1000-t_int]+args.cut_innercut[1000-t_int]))\n",
        "                for model_stat in model_stats:\n",
        "                 #errors on last step without +1, need to find source\n",
        "                    #when using SLIP Base model the dimensions need to be hard coded to avoid AttributeError: 'VisionTransformer' object has no attribute 'input_resolution'\n",
        "                    try:\n",
        "                        input_resolution=model_stat[\"clip_model\"].visual.input_resolution\n",
        "                    except:\n",
        "                        input_resolution=224\n",
        "\n",
        "                    cuts = MakeCutoutsES(input_resolution,\n",
        "                            randoffsetx = randomx,\n",
        "                            randoffsety = randomy,\n",
        "                            randsize = randomsize,\n",
        "                            randsize2 = randomsize2,\n",
        "                            scheduler = t_int,\n",
        "                            Overview = args.cut_overview[1000-t_int], \n",
        "                            InnerCrop = args.cut_innercut[1000-t_int],\n",
        "                            bucketcuts = args.bucket_cuts,\n",
        "                            IC_Size_Pow=args.cut_ic_pow[1000-t_int],\n",
        "                            IC_Grey = args.ic_gray_p[1000-t_int],\n",
        "                            overview_gray = args.ov_gray_p[1000-t_int],\n",
        "                            ov_proportion = args.percent_overview[1000-t_int],\n",
        "                            floor = args.floor_size[1000-t_int]\n",
        "                            )\n",
        "                    \n",
        "                    clip_in = normalize(cuts(x_in.add(1).div(2)))\n",
        "                    image_embeds = model_stat[\"clip_model\"].encode_image(clip_in).float()\n",
        "                    dists = spherical_dist_loss(image_embeds.unsqueeze(1), model_stat[\"target_embeds\"].unsqueeze(0))\n",
        "                    if (use_bucketpow or use_rand_bucketpow):\n",
        "                      dists = dists.view([args.bucket_cuts, n, -1])\n",
        "                    else:\n",
        "                      dists = dists.view([args.cut_overview[1000-t_int]+args.cut_innercut[1000-t_int], n, -1])\n",
        "                    if manual_weight and not (use_bucketpow or use_rand_bucketpow):\n",
        "                      dists[:args.cut_overview[1000-t_int]] *= args.overview_weight[1000-t_int]\n",
        "                    elif adaptive_weight:\n",
        "                      normalizer = 0\n",
        "                      if (use_bucketpow or use_rand_bucketpow):\n",
        "                        for i in range(args.bucket_cuts):\n",
        "                          dists[i] *= (cuts.sizes[i]/(224**2))**(args.cut_weight_multiplier[1000-t_int])\n",
        "                          normalizer += (cuts.sizes[i]/(224**2))**(args.cut_weight_multiplier[1000-t_int])\n",
        "                      else:\n",
        "                        for i in range((args.cut_innercut[1000-t_int])+(args.cut_overview[1000-t_int])):\n",
        "                          dists[i] *= (cuts.sizes[i]/(224**2))**(args.cut_weight_multiplier[1000-t_int]) #maybe should make clip_model size? but also maybe better to give greater weight to bigger models\n",
        "                          normalizer += (cuts.sizes[i]/(224**2))**(args.cut_weight_multiplier[1000-t_int])\n",
        "                    \n",
        "                    step_weights = []\n",
        "                    for i in range(0, len(model_stat[\"weights\"])):\n",
        "                      step_weights += [model_stat[\"weights\"][i][1000-t_int]]\n",
        "                      \n",
        "                    if np.absolute(np.sum(step_weights)) < 1e-3: raise RuntimeError('The weights must not sum to 0.')\n",
        "                    step_weights_divisor = np.absolute(np.sum(step_weights))\n",
        "                    step_weights = [item / step_weights_divisor for item in step_weights]\n",
        "                    step_weights = torch.tensor(step_weights, device=device)\n",
        "\n",
        "                    #losses = dists.mul(model_stat[\"weights\"]).sum(2).mean(0)\n",
        "                    losses = dists.mul(step_weights).sum(2).mean(0)\n",
        "                    if manual_weight and not (use_bucketpow or use_rand_bucketpow):\n",
        "                      if normalizeloss:\n",
        "                        losses = losses/(((args.cut_overview[1000-t_int])*(args.overview_weight[1000-t_int])+(args.cut_innercut[1000-t_int]))/((args.cut_innercut[1000-t_int])+(args.cut_overview[1000-t_int])))\n",
        "                    elif adaptive_weight:\n",
        "                      if (use_bucketpow or use_rand_bucketpow):\n",
        "                        losses = losses/(normalizer/args.bucket_cuts)         \n",
        "                      else:             \n",
        "                        losses = losses/(normalizer/((args.cut_innercut[1000-t_int])+(args.cut_overview[1000-t_int])))\n",
        "                    loss_values.append(losses.sum().item()) # log loss, probably shouldn't do per cutn_batch\n",
        "                    x_in_grad += torch.autograd.grad(losses.sum() * args.clip_guidance_scale[1000 - t_int], x_in)[0] / args.cutn_batches[1000 - t_int]\n",
        "              #range_losses = range_loss(out['pred_xstart'])\n",
        "              sat_losses = torch.abs(x_in - x_in.clamp(min=-1,max=1)).mean()\n",
        "              if not args.use_secondary_model[1000-t_int]:      \n",
        "                tv_losses = tv_loss(x_in)\n",
        "                if range_scale != 0:\n",
        "                    range_losses = range_loss(out['pred_xstart'])\n",
        "                    loss = tv_losses.sum() * tv_scale + range_losses.sum() * range_scale + sat_losses.sum() * sat_scale\n",
        "                else:\n",
        "                  loss = tv_losses.sum() * tv_scale + sat_losses.sum() * sat_scale\n",
        "              else:\n",
        "                loss = sat_losses.sum() * sat_scale\n",
        "              if (init is not None and init_scale and perlin_init is False):\n",
        "                  init_losses = lpips_model(x_in, init)\n",
        "                  loss = loss + init_losses.sum() * init_scale\n",
        "              if args.symmetry_loss and np.array(t.cpu())[0] > 10*args.symm_switch:\n",
        "                  sloss = symm_loss(x_in,lpips_model)\n",
        "                  loss = loss + sloss.sum() * args.sloss_scale\n",
        "              x_in_grad += torch.autograd.grad(loss, x_in)[0]\n",
        "              if torch.isnan(x_in_grad).any()==False:\n",
        "                  grad = -torch.autograd.grad(x_in, x, x_in_grad)[0]\n",
        "              else:\n",
        "                # print(\"NaN'd\")\n",
        "                x_is_NaN = True\n",
        "                grad = torch.zeros_like(x)\n",
        "          if args.clamp_grad and x_is_NaN == False:\n",
        "              magnitude = grad.square().mean().sqrt()\n",
        "              return grad * magnitude.clamp(max=args.clamp_max[1000 - t_int]) / magnitude  #min=-0.02, min=-clamp_max, \n",
        "          return grad      \n",
        "\n",
        "      if args.diffusion_sampling_mode == 'ddim':\n",
        "          sample_fn = diffusion.ddim_sample_loop_progressive\n",
        "      elif args.diffusion_sampling_mode == 'plms':\n",
        "          sample_fn = diffusion.plms_sample_loop_progressive\n",
        "      elif args.diffusion_sampling_mode == 'iplms':\n",
        "          sample_fn = diffusion.iplms_sample_loop_progressive\n",
        "      elif args.diffusion_sampling_mode == 'plms-ddim':\n",
        "          sample_fn = diffusion.PLMS_to_DDIM_sample_loop_progressive\n",
        "      elif args.diffusion_sampling_mode == 'iplms-ddim':\n",
        "          sample_fn = diffusion.IPLMS_to_DDIM_sample_loop_progressive\n",
        "\n",
        "      image_display = Output()\n",
        "      for i in range(args.n_batches):\n",
        "          if args.animation_mode == 'None':\n",
        "            display.clear_output(wait=True)\n",
        "            batchBar = tqdm(range(args.n_batches), desc =\"Batches\")\n",
        "            batchBar.n = i\n",
        "            batchBar.refresh()\n",
        "          print('')\n",
        "          display.display(image_display)\n",
        "          gc.collect()\n",
        "          torch.cuda.empty_cache()\n",
        "          cur_t = diffusion.num_timesteps - skip_steps - 1\n",
        "          total_steps = cur_t\n",
        "\n",
        "          if perlin_init:\n",
        "              init = regen_perlin()\n",
        "\n",
        "          if args.diffusion_sampling_mode == 'ddim':\n",
        "              samples = sample_fn(\n",
        "                  model,\n",
        "                  (batch_size, 3, side_y, side_x),\n",
        "                  clip_denoised=clip_denoised,\n",
        "                  model_kwargs={},\n",
        "                  cond_fn=cond_fn,\n",
        "                  progress=True,\n",
        "                  skip_timesteps=skip_steps,\n",
        "                  init_image=init,\n",
        "                  randomize_class=randomize_class,\n",
        "                  eta=eta,\n",
        "                  transformation_fn=resize_transformation_fn,\n",
        "                  transformation_percent=args.transformation_percent\n",
        "              )\n",
        "          elif args.diffusion_sampling_mode == 'plms':\n",
        "              samples = sample_fn(\n",
        "                  model,\n",
        "                  (batch_size, 3, side_y, side_x),\n",
        "                  clip_denoised=clip_denoised,\n",
        "                  model_kwargs={},\n",
        "                  cond_fn=cond_fn,\n",
        "                  progress=True,\n",
        "                  skip_timesteps=skip_steps,\n",
        "                  init_image=init,\n",
        "                  randomize_class=randomize_class,\n",
        "                  order=2,\n",
        "              )\n",
        "          elif args.diffusion_sampling_mode == 'iplms':\n",
        "              samples = sample_fn(\n",
        "                  model,\n",
        "                  (batch_size, 3, side_y, side_x),\n",
        "                  clip_denoised=clip_denoised,\n",
        "                  model_kwargs={},\n",
        "                  cond_fn=cond_fn,\n",
        "                  progress=True,\n",
        "                  skip_timesteps=skip_steps,\n",
        "                  init_image=init,\n",
        "                  randomize_class=randomize_class,\n",
        "                  order=2,\n",
        "              )\n",
        "          else:\n",
        "              samples = sample_fn(\n",
        "                  model,\n",
        "                  (batch_size, 3, side_y, side_x),\n",
        "                  clip_denoised=clip_denoised,\n",
        "                  model_kwargs={},\n",
        "                  cond_fn=cond_fn,\n",
        "                  progress=True,\n",
        "                  switch_timestep=switch_timestep,\n",
        "                  skip_timesteps=skip_steps,\n",
        "                  init_image=init,\n",
        "                  randomize_class=randomize_class,\n",
        "                  order=2,\n",
        "              )\n",
        "          \n",
        "          \n",
        "          # with run_display:\n",
        "          # display.clear_output(wait=True)\n",
        "          for j, sample in enumerate(samples):    \n",
        "            cur_t -= 1\n",
        "            intermediateStep = False\n",
        "            if args.steps_per_checkpoint is not None:\n",
        "                if j % steps_per_checkpoint == 0 and j > 0:\n",
        "                  intermediateStep = True\n",
        "            elif j in args.intermediate_saves:\n",
        "              intermediateStep = True\n",
        "            with image_display:\n",
        "              if j % args.display_rate == 0 or cur_t == -1 or intermediateStep == True:\n",
        "                  if prompt_in_filename and animation_mode == \"None\":\n",
        "                    tempname = str(frame_prompt[0])\n",
        "                    tempname = tempname.replace(\" \", \"_\")\n",
        "                    name = tempname[:42]\n",
        "                  else:\n",
        "                    name = args.batch_name\n",
        "                  for k, image in enumerate(sample['pred_xstart']):\n",
        "                      # tqdm.write(f'Batch {i}, step {j}, output {k}:')\n",
        "                      current_time = datetime.now().strftime('%y%m%d-%H%M%S_%f')\n",
        "                      percent = math.ceil(j/total_steps*100)\n",
        "                      if args.n_batches > 0:\n",
        "                        #if intermediates are saved to the subfolder, don't append a step or percentage to the name\n",
        "                        if cur_t == -1 and args.intermediates_in_subfolder is True:\n",
        "                          save_num = f'{frame_num:04}' if animation_mode != \"None\" else i\n",
        "                          filename = f'{name}({args.batchNum})_{save_num}.png'\n",
        "                        else:\n",
        "                          #If we're working with percentages, append it\n",
        "                          if args.steps_per_checkpoint is not None:\n",
        "                            filename = f'{name}({args.batchNum})_{i:04}-{percent:02}%.png'\n",
        "                          # Or else, iIf we're working with specific steps, append those\n",
        "                          else:\n",
        "                            filename = f'{name}({args.batchNum})_{i:04}.png'\n",
        "                      image = TF.to_pil_image(image.add(1).div(2).clamp(0, 1))\n",
        "                      if j % args.display_rate == 0 or cur_t == -1:\n",
        "                        image.save('progress.png')\n",
        "                        display.clear_output(wait=True)\n",
        "                        display.display(display.Image('progress.png'))\n",
        "                      if args.steps_per_checkpoint is not None:\n",
        "                        if j % args.steps_per_checkpoint == 0 and j > 0:\n",
        "                          if args.intermediates_in_subfolder is True:\n",
        "                            image.save(f'{partialFolder}/{filename}')\n",
        "                          else:\n",
        "                            image.save(f'{batchFolder}/{filename}')\n",
        "                      else:\n",
        "                        if j in args.intermediate_saves:\n",
        "                          if args.intermediates_in_subfolder is True:\n",
        "                            image.save(f'{partialFolder}/{filename}')\n",
        "                          else:\n",
        "                            image.save(f'{batchFolder}/{filename}')\n",
        "                      if cur_t == -1:\n",
        "                        if frame_num == 0:\n",
        "                          try:\n",
        "                            save_settings()\n",
        "                          except:\n",
        "                            print('Could not save settings, oh well...')\n",
        "                        if args.animation_mode != \"None\":\n",
        "                          image.save('prevFrame.png')\n",
        "                        image.save(f'{batchFolder}/{filename}')\n",
        "                        if args.animation_mode == \"None\":\n",
        "                          print('Incrementing seed by one.')\n",
        "                          seed = seed + 1\n",
        "                          np.random.seed(seed)\n",
        "                          random.seed(seed)\n",
        "                          torch.manual_seed(seed)\n",
        "                          torch.cuda.manual_seed_all(seed)\n",
        "                          torch.backends.cudnn.deterministic = True\n",
        "                        if args.animation_mode == \"3D\":\n",
        "                          # If turbo, save a blended image\n",
        "                          if turbo_mode and frame_num > 0:\n",
        "                            # Mix new image with prevFrameScaled\n",
        "                            blend_factor = (1)/int(turbo_steps)\n",
        "                            newFrame = cv2.imread('prevFrame.png') # This is already updated..\n",
        "                            prev_frame_warped = cv2.imread('prevFrameScaled.png')\n",
        "                            blendedImage = cv2.addWeighted(newFrame, blend_factor, prev_frame_warped, (1-blend_factor), 0.0)\n",
        "                            cv2.imwrite(f'{batchFolder}/{filename}',blendedImage)\n",
        "                          else:\n",
        "                            image.save(f'{batchFolder}/{filename}')\n",
        "\n",
        "                          if vr_mode:\n",
        "                            generate_eye_views(TRANSLATION_SCALE, batchFolder, filename, frame_num, midas_model, midas_transform)\n",
        "\n",
        "                        # if frame_num != args.max_frames-1:\n",
        "                        #   display.clear_output()\n",
        "          \n",
        "          plt.plot(np.array(loss_values), 'r')\n",
        "\n",
        "def generate_eye_views(trans_scale,batchFolder,filename,frame_num,midas_model, midas_transform):\n",
        "   for i in range(2):\n",
        "      theta = vr_eye_angle * (math.pi/180)\n",
        "      ray_origin = math.cos(theta) * vr_ipd / 2 * (-1.0 if i==0 else 1.0)\n",
        "      ray_rotation = (theta if i==0 else -theta)\n",
        "      translate_xyz = [-(ray_origin)*trans_scale, 0,0]\n",
        "      rotate_xyz = [0, (ray_rotation), 0]\n",
        "      rot_mat = p3dT.euler_angles_to_matrix(torch.tensor(rotate_xyz, device=device), \"XYZ\").unsqueeze(0)\n",
        "      transformed_image = dxf.transform_image_3d(f'{batchFolder}/{filename}', midas_model, midas_transform, DEVICE,\n",
        "                                                      rot_mat, translate_xyz, args.near_plane, args.far_plane,\n",
        "                                                      args.fov, padding_mode=args.padding_mode,\n",
        "                                                      sampling_mode=args.sampling_mode, midas_weight=args.midas_weight,spherical=True)\n",
        "      eye_file_path = batchFolder+f\"/frame_{frame_num:04}\" + ('_l' if i==0 else '_r')+'.png'\n",
        "      transformed_image.save(eye_file_path)\n",
        "\n",
        "def save_settings():\n",
        "    setting_list = {\n",
        "      'text_prompts': text_prompts,\n",
        "      'image_prompts': image_prompts,\n",
        "      'clip_guidance_scale': str(clip_guidance_scale),\n",
        "      'tv_scale': tv_scale,\n",
        "      'range_scale': range_scale,\n",
        "      'sat_scale': sat_scale,\n",
        "      'symmetry_loss':symmetry_loss,\n",
        "      'sloss_scale':symm_loss_scale,\n",
        "      'symm_switch':symm_switch,\n",
        "      'cutn_batches': str(cutn_batches),\n",
        "      'max_frames': max_frames,\n",
        "      'interp_spline': interp_spline,\n",
        "      'init_image': init_image,\n",
        "      'init_scale': init_scale,\n",
        "      'skip_steps': skip_steps,\n",
        "      'frames_scale': frames_scale,\n",
        "      'frames_skip_steps': frames_skip_steps,\n",
        "      'perlin_init': perlin_init,\n",
        "      'perlin_mode': perlin_mode,\n",
        "      'skip_augs': skip_augs,\n",
        "      'randomize_class': randomize_class,\n",
        "      'clip_denoised': clip_denoised,\n",
        "      'clamp_grad': clamp_grad,\n",
        "      'clamp_max': str(clamp_max),\n",
        "      'seed': seed,\n",
        "      'fuzzy_prompt': fuzzy_prompt,\n",
        "      'rand_mag': rand_mag,\n",
        "      'eta': eta,\n",
        "      'width': width_height[0],\n",
        "      'height': width_height[1],\n",
        "      'diffusion_model': diffusion_model,\n",
        "      'use_secondary_model': str(use_secondary_model),\n",
        "      'steps': steps,\n",
        "      'diffusion_steps': diffusion_steps,\n",
        "      'diffusion_sampling_mode': diffusion_sampling_mode,\n",
        "      'ViTB32': ViTB32,\n",
        "      'ViTB16': ViTB16,\n",
        "      'ViTL14': ViTL14,\n",
        "      'ViTL14_336px': ViTL14_336px,\n",
        "      'RN101': RN101,\n",
        "      'RN50': RN50,\n",
        "      'RN50x4': RN50x4,\n",
        "      'RN50x16': RN50x16,\n",
        "      'RN50x64': RN50x64,\n",
        "      'ViTB32_laion2b_e16': ViTB32_laion2b_e16,\n",
        "      'ViTB32_laion400m_e31': ViTB32_laion400m_e31,\n",
        "      'ViTB32_laion400m_32': ViTB32_laion400m_32,\n",
        "      'ViTB32quickgelu_laion400m_e31': ViTB32quickgelu_laion400m_e31,\n",
        "      'ViTB32quickgelu_laion400m_e32': ViTB32quickgelu_laion400m_e32,\n",
        "      'ViTB16_laion400m_e31': ViTB16_laion400m_e31,\n",
        "      'ViTB16_laion400m_e32': ViTB16_laion400m_e32,\n",
        "      'RN50_yffcc15m': RN50_yffcc15m,\n",
        "      'RN50_cc12m': RN50_cc12m,\n",
        "      'RN50_quickgelu_yfcc15m': RN50_quickgelu_yfcc15m,\n",
        "      'RN50_quickgelu_cc12m': RN50_quickgelu_cc12m,\n",
        "      'RN101_yfcc15m': RN101_yfcc15m,\n",
        "      'RN101_quickgelu_yfcc15m': RN101_quickgelu_yfcc15m,\n",
        "      'cut_overview': str(cut_overview),\n",
        "      'cut_innercut': str(cut_innercut),\n",
        "      'cut_ic_pow': str(cut_ic_pow),\n",
        "      'cut_icgray_p': str(ic_gray_p),\n",
        "      'ov_gray': str(ov_gray_p),\n",
        "      'percent_overview': str(percent_overview),\n",
        "      'overview_weight': str(overview_weight),\n",
        "      'floor': str(floor_size),\n",
        "      'user_floor': use_floor,\n",
        "      'adaptive_weight': adaptive_weight,\n",
        "      'cut_weight_multiplier': str(cut_weight_multiplier),\n",
        "      'ceiling': ceiling,\n",
        "      'all_models_same_cuts': all_models_same_cuts,\n",
        "      'use_floor': use_floor,\n",
        "      'key_frames': key_frames,\n",
        "      'max_frames': max_frames,\n",
        "      'angle': angle,\n",
        "      'zoom': zoom,\n",
        "      'translation_x': translation_x,\n",
        "      'translation_y': translation_y,\n",
        "      'translation_z': translation_z,\n",
        "      'rotation_3d_x': rotation_3d_x,\n",
        "      'rotation_3d_y': rotation_3d_y,\n",
        "      'rotation_3d_z': rotation_3d_z,\n",
        "      'midas_depth_model': midas_depth_model,\n",
        "      'midas_weight': midas_weight,\n",
        "      'near_plane': near_plane,\n",
        "      'far_plane': far_plane,\n",
        "      'fov': fov,\n",
        "      'padding_mode': padding_mode,\n",
        "      'sampling_mode': sampling_mode,\n",
        "      'video_init_path':video_init_path,\n",
        "      'extract_nth_frame':extract_nth_frame,\n",
        "      'video_init_seed_continuity': video_init_seed_continuity,\n",
        "      'turbo_mode':turbo_mode,\n",
        "      'turbo_steps':turbo_steps,\n",
        "      'turbo_preroll':turbo_preroll,\n",
        "      'use_horizontal_symmetry':use_horizontal_symmetry,\n",
        "      'use_vertical_symmetry':use_vertical_symmetry,\n",
        "      'transformation_percent':transformation_percent,\n",
        "      #video init settings\n",
        "      'video_init_steps': video_init_steps,\n",
        "      'video_init_clip_guidance_scale': video_init_clip_guidance_scale,\n",
        "      'video_init_tv_scale': video_init_tv_scale,\n",
        "      'video_init_range_scale': video_init_range_scale,\n",
        "      'video_init_sat_scale': video_init_sat_scale,\n",
        "      'video_init_cutn_batches': video_init_cutn_batches,\n",
        "      'video_init_skip_steps': video_init_skip_steps,\n",
        "      'video_init_frames_scale': video_init_frames_scale,\n",
        "      'video_init_frames_skip_steps': video_init_frames_skip_steps,\n",
        "      #warp settings\n",
        "      'video_init_flow_warp':video_init_flow_warp,\n",
        "      'video_init_flow_blend':video_init_flow_blend,\n",
        "      'video_init_check_consistency':video_init_check_consistency,\n",
        "      'video_init_blend_mode':video_init_blend_mode,\n",
        "      'noise_injector': noise_injector,\n",
        "      'noise_injector_threshold': noise_injector_threshold,\n",
        "      'noise_injector_px_size': noise_injector_px_size,\n",
        "      'noise_injector_sigma': noise_injector_sigma,\n",
        "      'noise_injector_spice': noise_injector_spice,\n",
        "      'noise_injector_mask_feather': noise_injector_mask_feather,\n",
        "      'copy_palette': copy_palette,\n",
        "      'copy_palette_image': copy_palette_image,\n",
        "    }\n",
        "    # print('Settings:', setting_list)\n",
        "    with open(f\"{batchFolder}/{batch_name}({batchNum})_settings.txt\", \"w+\", encoding=\"utf-8\") as f:   #save settings\n",
        "        json.dump(setting_list, f, ensure_ascii=False, indent=4)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "source": [
        "#@title 1.7 functions for zippy's animation toolbox\n",
        "def soft_limit(x):\n",
        "  #zippy soft limiter\n",
        "  #value from -n to n, always return a value -1<x<1\n",
        "  #soft_limiter_knee set in params = 0.97 # where does compression start?\n",
        "  soft_sign = x/abs(x)\n",
        "  soft_overage = ((abs(x)-soft_limiter_knee)+(abs(abs(x)-soft_limiter_knee)))/2\n",
        "  soft_base = abs(x)-soft_overage\n",
        "  soft_limited_x = soft_base + torch.tanh(soft_overage/(1-soft_limiter_knee))*(1-soft_limiter_knee)\n",
        "  return soft_limited_x*soft_sign\n",
        "\n",
        "def center_crop(img,crop_pct):\n",
        "  #zippy color kit\n",
        "  dimensions = img.shape \n",
        "  # height, width, number of channels in image\n",
        "  height = img.shape[0]\n",
        "  width = img.shape[1]\n",
        "  channels = img.shape[2]\n",
        "  h1=int((height/2)*(1-crop_pct))\n",
        "  h2=int((height/2)*(1+crop_pct))\n",
        "  w1=int((width/2)*(1-crop_pct))\n",
        "  w2=int((width/2)*(1+crop_pct))\n",
        "  crop_img = img[h1:h2, w1:w2]\n",
        "  return crop_img\n",
        "\n",
        "def get_edge_img(img):\n",
        "  #zippy color kit\n",
        "  canny_min = .66*img.mean()\n",
        "  canny_max = 1.33*img.mean()  \n",
        "  get_edge_img = cv2.Canny(img,canny_min,canny_max,True) \n",
        "  return get_edge_img\n",
        "\n",
        "def get_edge_mix(img):\n",
        "  #zippy color kit\n",
        "  edges = get_edge_img(img)\n",
        "  edge_mix = edges.mean()\n",
        "  return edge_mix\n",
        "\n",
        "def calc_contrast(img):\n",
        "  #from https://stackoverflow.com/questions/58821130/how-to-calculate-the-contrast-of-an-image\n",
        "  img_grey = cv2.cvtColor((img*255).astype(np.uint8), cv2.COLOR_BGR2GRAY)\n",
        "  contrast = img_grey.std()\n",
        "  return contrast\n",
        "\n",
        "def calc_brightness(img):\n",
        "  #zippy color kit\n",
        "  #from https://stackoverflow.com/questions/58821130/how-to-calculate-the-contrast-of-an-image\n",
        "  img_grey = cv2.cvtColor((img*255).astype(np.uint8), cv2.COLOR_BGR2GRAY)\n",
        "  brightness = img_grey.mean()\n",
        "  return brightness\n",
        "\n",
        "def contrast_stretch(img):\n",
        "  #zippy color kit\n",
        "  #requires a cv2 Image (val 0-1) \n",
        "  #also https://stackoverflow.com/questions/64484423/cv2-lut-throws-an-error-when-performing-gamma-correction-assertion-failed\n",
        "  xp = [0, 64, 128, 192, 255]\n",
        "  fp = [0, 16, 128, 240, 255]\n",
        "  x = np.arange(256)\n",
        "  table = np.interp(x, xp, fp).astype('uint8')\n",
        "  #img = cv2.LUT((255*img).astype(np.uint8), table)\n",
        "  #img = (img.astype(np.float)) / 255\n",
        "  img = cv2.LUT((img).astype(np.uint8), table)\n",
        "  img = (img.astype(np.float))\n",
        "  return img\n",
        "\n",
        "def contrast_stretch_lite(img):\n",
        "  #zippy color kit\n",
        "  xp = [0, 64, 128, 192, 255]\n",
        "  fp = [0, 32, 128, 224, 255]\n",
        "  x = np.arange(256)\n",
        "  table = np.interp(x, xp, fp).astype('uint8')\n",
        "  #img = cv2.LUT((255*img).astype(np.uint8), table)\n",
        "  #img = (img.astype(np.float)) / 255\n",
        "  img = cv2.LUT((img).astype(np.uint8), table)\n",
        "  img = (img.astype(np.float))\n",
        "  return img\n",
        "\n",
        "def add_noise(img,noise_mix,noise_injector_px_size,\n",
        "              noise_injector_spice,noise_injector_mask_feather,\n",
        "              noise_injector_sigma):\n",
        "  # zippy noise injector csa\n",
        "  # make edge map from edge image\n",
        "  width = int(img.shape[1])\n",
        "  height = int(img.shape[0])\n",
        "  dim_hw = (height, width)\n",
        "  dim_wh = (width, height)\n",
        "\n",
        "  dim = (int(img.shape[1]), int(img.shape[0]))\n",
        "  edge_img = get_edge_img(img)\n",
        "  kernel = np.ones((7,7), np.uint8)\n",
        "  kernel2 = np.ones((21,21), np.uint8)\n",
        "  edge_img_dilation = cv2.dilate(edge_img, kernel, iterations=3)\n",
        "  edge_img_dilation2 = cv2.dilate(edge_img, kernel2, iterations=3)\n",
        "  blended_dilation = (edge_img_dilation/2 + edge_img_dilation2/2)\n",
        "  blur_size = noise_injector_mask_feather*2+1\n",
        "  \n",
        "  blur_kernel = (blur_size,blur_size)\n",
        "  #3 blurs, just for grins\n",
        "  edge_blurred_img = cv2.GaussianBlur(blended_dilation, blur_kernel, 0)\n",
        "  edge_blurred_img = cv2.GaussianBlur(edge_blurred_img, blur_kernel, 0)\n",
        "  edge_blurred_img = cv2.GaussianBlur(edge_blurred_img, blur_kernel, 0)\n",
        "  mask = (255-edge_blurred_img)\n",
        "  mask = mask.reshape(*mask.shape, 1)/255\n",
        "\n",
        "  #make gaussian mask\n",
        "  #https://stackoverflow.com/questions/55261087/adding-gaussian-noise-to-image\n",
        "  mean = 0\n",
        "  #var = 100\n",
        "  #sigma = var ** 0.5\n",
        "  sigma = noise_injector_sigma\n",
        "  #sigma = 64 #larger numbers = bigger blobs\n",
        "  #noise image is expecting dim_hw\n",
        "  gaussian = np.random.normal(mean, sigma, dim_hw) # dim must be (H,W) \n",
        "  gaus_blur_kernel = (15,15)\n",
        "  blurred_gaussian = cv2.GaussianBlur(gaussian, gaus_blur_kernel, 0)\n",
        "  gaussian_mask = blurred_gaussian.reshape(*blurred_gaussian.shape, 1)\n",
        "  mask = np.clip(mask, 0, 1)\n",
        "  gaussian_mask = np.clip(gaussian_mask, 0, 1)\n",
        "  combo_mask = mask*gaussian_mask\n",
        "\n",
        "  #now make a noise pattern\n",
        "  noise_img =  np.random.normal(loc=0, scale=1, size=img.shape)\n",
        "  noise_img *= 255\n",
        "  #change pixel density...\n",
        "  noise_px_size = noise_injector_px_size  \n",
        "  noise_img = center_crop(noise_img,1/noise_px_size)\n",
        "  #modify noise color mix to original img\n",
        "  colored_noise = exposure.match_histograms(noise_img, img, multichannel=True)\n",
        "  #blend in 'spice' \n",
        "  colored_noise = noise_img * noise_injector_spice + colored_noise * (1-noise_injector_spice)  \n",
        "  \n",
        "  #resize to orig.. using dim from above\n",
        "  #cv2.resize is expecting dim_wh\n",
        "  colored_noise = cv2.resize(colored_noise, dim_wh, interpolation = cv2.INTER_AREA) \n",
        "  noise_blur_size = noise_px_size * 2 + 1\n",
        "  noise_blur_kernel = (noise_blur_size,noise_blur_size)    \n",
        "  colored_noise = cv2.GaussianBlur(colored_noise, noise_blur_kernel, 0)\n",
        "  colored_noise = cv2.GaussianBlur(colored_noise, noise_blur_kernel, 0) \n",
        "  \n",
        "  masked_noise = colored_noise * combo_mask\n",
        "  scaled_mask = combo_mask * noise_mix\n",
        "  #blend in noise using mask\n",
        "  final_img = colored_noise * scaled_mask + img * (1 - scaled_mask)\n",
        "\n",
        "  #uncomment below to see elements of noise injector\n",
        "  #cv2_imshow(edge_img)\n",
        "  #cv2_imshow(blended_dilation)\n",
        "  #cv2_imshow(mask*255)#mask is 0-1\n",
        "  #cv2_imshow(gaussian_mask*255)\n",
        "  #cv2_imshow(combo_mask*255)\n",
        "  #cv2_imshow(noise_img)\n",
        "  #cv2_imshow(colored_noise)\n",
        "  #cv2_imshow(final_img)\n",
        "  #print('corrected contrast is ',(calc_contrast(final_img)))\n",
        "  #print(calc_contrast(final_img))\n",
        "  return final_img\n"
      ],
      "metadata": {
        "cellView": "form",
        "id": "U9Fa_ztVo9HK"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "cellView": "form",
        "id": "DefSecModel"
      },
      "source": [
        "#@title 1.6 Define the secondary diffusion model\n",
        "\n",
        "def append_dims(x, n):\n",
        "    return x[(Ellipsis, *(None,) * (n - x.ndim))]\n",
        "\n",
        "\n",
        "def expand_to_planes(x, shape):\n",
        "    return append_dims(x, len(shape)).repeat([1, 1, *shape[2:]])\n",
        "\n",
        "\n",
        "def alpha_sigma_to_t(alpha, sigma):\n",
        "    return torch.atan2(sigma, alpha) * 2 / math.pi\n",
        "\n",
        "\n",
        "def t_to_alpha_sigma(t):\n",
        "    return torch.cos(t * math.pi / 2), torch.sin(t * math.pi / 2)\n",
        "\n",
        "\n",
        "@dataclass\n",
        "class DiffusionOutput:\n",
        "    v: torch.Tensor\n",
        "    pred: torch.Tensor\n",
        "    eps: torch.Tensor\n",
        "\n",
        "\n",
        "class ConvBlock(nn.Sequential):\n",
        "    def __init__(self, c_in, c_out):\n",
        "        super().__init__(\n",
        "            nn.Conv2d(c_in, c_out, 3, padding=1),\n",
        "            nn.ReLU(inplace=True),\n",
        "        )\n",
        "\n",
        "\n",
        "class SkipBlock(nn.Module):\n",
        "    def __init__(self, main, skip=None):\n",
        "        super().__init__()\n",
        "        self.main = nn.Sequential(*main)\n",
        "        self.skip = skip if skip else nn.Identity()\n",
        "\n",
        "    def forward(self, input):\n",
        "        return torch.cat([self.main(input), self.skip(input)], dim=1)\n",
        "\n",
        "\n",
        "class FourierFeatures(nn.Module):\n",
        "    def __init__(self, in_features, out_features, std=1.):\n",
        "        super().__init__()\n",
        "        assert out_features % 2 == 0\n",
        "        self.weight = nn.Parameter(torch.randn([out_features // 2, in_features]) * std)\n",
        "\n",
        "    def forward(self, input):\n",
        "        f = 2 * math.pi * input @ self.weight.T\n",
        "        return torch.cat([f.cos(), f.sin()], dim=-1)\n",
        "\n",
        "\n",
        "class SecondaryDiffusionImageNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super().__init__()\n",
        "        c = 64  # The base channel count\n",
        "\n",
        "        self.timestep_embed = FourierFeatures(1, 16)\n",
        "\n",
        "        self.net = nn.Sequential(\n",
        "            ConvBlock(3 + 16, c),\n",
        "            ConvBlock(c, c),\n",
        "            SkipBlock([\n",
        "                nn.AvgPool2d(2),\n",
        "                ConvBlock(c, c * 2),\n",
        "                ConvBlock(c * 2, c * 2),\n",
        "                SkipBlock([\n",
        "                    nn.AvgPool2d(2),\n",
        "                    ConvBlock(c * 2, c * 4),\n",
        "                    ConvBlock(c * 4, c * 4),\n",
        "                    SkipBlock([\n",
        "                        nn.AvgPool2d(2),\n",
        "                        ConvBlock(c * 4, c * 8),\n",
        "                        ConvBlock(c * 8, c * 4),\n",
        "                        nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False),\n",
        "                    ]),\n",
        "                    ConvBlock(c * 8, c * 4),\n",
        "                    ConvBlock(c * 4, c * 2),\n",
        "                    nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False),\n",
        "                ]),\n",
        "                ConvBlock(c * 4, c * 2),\n",
        "                ConvBlock(c * 2, c),\n",
        "                nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False),\n",
        "            ]),\n",
        "            ConvBlock(c * 2, c),\n",
        "            nn.Conv2d(c, 3, 3, padding=1),\n",
        "        )\n",
        "\n",
        "    def forward(self, input, t):\n",
        "        timestep_embed = expand_to_planes(self.timestep_embed(t[:, None]), input.shape)\n",
        "        v = self.net(torch.cat([input, timestep_embed], dim=1))\n",
        "        alphas, sigmas = map(partial(append_dims, n=v.ndim), t_to_alpha_sigma(t))\n",
        "        pred = input * alphas - v * sigmas\n",
        "        eps = input * sigmas + v * alphas\n",
        "        return DiffusionOutput(v, pred, eps)\n",
        "\n",
        "\n",
        "class SecondaryDiffusionImageNet2(nn.Module):\n",
        "    def __init__(self):\n",
        "        super().__init__()\n",
        "        c = 64  # The base channel count\n",
        "        cs = [c, c * 2, c * 2, c * 4, c * 4, c * 8]\n",
        "\n",
        "        self.timestep_embed = FourierFeatures(1, 16)\n",
        "        self.down = nn.AvgPool2d(2)\n",
        "        self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False)\n",
        "\n",
        "        self.net = nn.Sequential(\n",
        "            ConvBlock(3 + 16, cs[0]),\n",
        "            ConvBlock(cs[0], cs[0]),\n",
        "            SkipBlock([\n",
        "                self.down,\n",
        "                ConvBlock(cs[0], cs[1]),\n",
        "                ConvBlock(cs[1], cs[1]),\n",
        "                SkipBlock([\n",
        "                    self.down,\n",
        "                    ConvBlock(cs[1], cs[2]),\n",
        "                    ConvBlock(cs[2], cs[2]),\n",
        "                    SkipBlock([\n",
        "                        self.down,\n",
        "                        ConvBlock(cs[2], cs[3]),\n",
        "                        ConvBlock(cs[3], cs[3]),\n",
        "                        SkipBlock([\n",
        "                            self.down,\n",
        "                            ConvBlock(cs[3], cs[4]),\n",
        "                            ConvBlock(cs[4], cs[4]),\n",
        "                            SkipBlock([\n",
        "                                self.down,\n",
        "                                ConvBlock(cs[4], cs[5]),\n",
        "                                ConvBlock(cs[5], cs[5]),\n",
        "                                ConvBlock(cs[5], cs[5]),\n",
        "                                ConvBlock(cs[5], cs[4]),\n",
        "                                self.up,\n",
        "                            ]),\n",
        "                            ConvBlock(cs[4] * 2, cs[4]),\n",
        "                            ConvBlock(cs[4], cs[3]),\n",
        "                            self.up,\n",
        "                        ]),\n",
        "                        ConvBlock(cs[3] * 2, cs[3]),\n",
        "                        ConvBlock(cs[3], cs[2]),\n",
        "                        self.up,\n",
        "                    ]),\n",
        "                    ConvBlock(cs[2] * 2, cs[2]),\n",
        "                    ConvBlock(cs[2], cs[1]),\n",
        "                    self.up,\n",
        "                ]),\n",
        "                ConvBlock(cs[1] * 2, cs[1]),\n",
        "                ConvBlock(cs[1], cs[0]),\n",
        "                self.up,\n",
        "            ]),\n",
        "            ConvBlock(cs[0] * 2, cs[0]),\n",
        "            nn.Conv2d(cs[0], 3, 3, padding=1),\n",
        "        )\n",
        "\n",
        "    def forward(self, input, t):\n",
        "        timestep_embed = expand_to_planes(self.timestep_embed(t[:, None]), input.shape)\n",
        "        v = self.net(torch.cat([input, timestep_embed], dim=1))\n",
        "        alphas, sigmas = map(partial(append_dims, n=v.ndim), t_to_alpha_sigma(t))\n",
        "        pred = input * alphas - v * sigmas\n",
        "        eps = input * sigmas + v * alphas\n",
        "        return DiffusionOutput(v, pred, eps)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DiffClipSetTop"
      },
      "source": [
        "# 2. Diffusion and CLIP model settings"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ModelSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Models Settings (note: For pixel art, the best is pixelartdiffusion_expanded):**\n",
        "diffusion_model = \"512x512_diffusion_uncond_finetune_008100\" #@param [\"256x256_diffusion_uncond\", \"512x512_diffusion_uncond_finetune_008100\", \"portrait_generator_v001\", \"pixelartdiffusion_expanded\", \"pixel_art_diffusion_hard_256\", \"pixel_art_diffusion_soft_256\", \"pixelartdiffusion4k\", \"watercolordiffusion_2\", \"watercolordiffusion\", \"PulpSciFiDiffusion\", \"custom\"]\n",
        "\n",
        "use_secondary_model = \"[False]*400+[False]*600\" #@param {type: 'string'}\n",
        "diffusion_sampling_mode = 'ddim' #@param ['ddim','plms','iplms','plms-ddim','iplms-ddim']\n",
        "switch_timestep = 80 #@param {type:'integer'}\n",
        "\n",
        "#switch_timestep =  89#@param{type: 'raw'}\n",
        "#@markdown #####**Custom model:**\n",
        "custom_path = '/content/drive/MyDrive/deep_learning/ddpm/ema_0.9999_058000.pt'#@param {type: 'string'}\n",
        "\n",
        "#@markdown #####**CLIP settings:**\n",
        "use_checkpoint = True #param {type: 'boolean'}\n",
        "ViTB32 = True #@param{type:\"boolean\"}\n",
        "ViTB16 = True #@param{type:\"boolean\"}\n",
        "ViTL14 = False #@param{type:\"boolean\"}\n",
        "ViTL14_336px = False #@param{type:\"boolean\"}\n",
        "RN101 = False #@param{type:\"boolean\"}\n",
        "RN50 = True #@param{type:\"boolean\"}\n",
        "RN50x4 = False #@param{type:\"boolean\"}\n",
        "RN50x16 = False #@param{type:\"boolean\"}\n",
        "RN50x64 = False #@param{type:\"boolean\"}\n",
        "\n",
        "#@markdown #####**OpenCLIP settings:** openclip models can still be used, but have been commented out to take less space\n",
        "ViTB32_laion2b_e16 = False #param{type:\"boolean\"}\n",
        "ViTB32_laion400m_e31 = False #param{type:\"boolean\"}\n",
        "ViTB32_laion400m_32 = False #param{type:\"boolean\"}\n",
        "ViTB32quickgelu_laion400m_e31 = False #param{type:\"boolean\"}\n",
        "ViTB32quickgelu_laion400m_e32 = False #param{type:\"boolean\"}\n",
        "ViTB16_laion400m_e31 = False #param{type:\"boolean\"}\n",
        "ViTB16_laion400m_e32 = False #param{type:\"boolean\"}\n",
        "RN50_yffcc15m = False #param{type:\"boolean\"}\n",
        "RN50_cc12m = False #param{type:\"boolean\"}\n",
        "RN50_quickgelu_yfcc15m = False #param{type:\"boolean\"}\n",
        "RN50_quickgelu_cc12m = False #param{type:\"boolean\"}\n",
        "RN101_yfcc15m = False #param{type:\"boolean\"}\n",
        "RN101_quickgelu_yfcc15m = False #param{type:\"boolean\"}\n",
        "\n",
        "VITL14open = False #param{type:\"boolean\"}\n",
        "VITH14open = False #param{type:\"boolean\"}\n",
        "VITG14open = False #param{type:\"boolean\"}\n",
        "\n",
        "#@markdown If you're having issues with model downloads, check this to compare SHA's:\n",
        "check_model_SHA = False #@param{type:\"boolean\"}\n",
        "\n",
        "diff_model_map = {\n",
        "    '256x256_diffusion_uncond': { 'downloaded': False, 'sha': 'a37c32fffd316cd494cf3f35b339936debdc1576dad13fe57c42399a5dbc78b1', 'uri_list': ['https://openaipublic.blob.core.windows.net/diffusion/jul-2021/256x256_diffusion_uncond.pt', 'https://www.dropbox.com/s/9tqnqo930mpnpcn/256x256_diffusion_uncond.pt'] },\n",
        "    '512x512_diffusion_uncond_finetune_008100': { 'downloaded': False, 'sha': '9c111ab89e214862b76e1fa6a1b3f1d329b1a88281885943d2cdbe357ad57648', 'uri_list': ['https://huggingface.co/lowlevelware/512x512_diffusion_unconditional_ImageNet/resolve/main/512x512_diffusion_uncond_finetune_008100.pt', 'https://the-eye.eu/public/AI/models/512x512_diffusion_unconditional_ImageNet/512x512_diffusion_uncond_finetune_008100.pt'] },\n",
        "    'portrait_generator_v001': { 'downloaded': False, 'sha': 'b7e8c747af880d4480b6707006f1ace000b058dd0eac5bb13558ba3752d9b5b9', 'uri_list': ['https://huggingface.co/felipe3dartist/portrait_generator_v001/resolve/main/portrait_generator_v001_ema_0.9999_1MM.pt'] },\n",
        "    'pixelartdiffusion_expanded': { 'downloaded': False, 'sha': 'a73b40556634034bf43b5a716b531b46fb1ab890634d854f5bcbbef56838739a', 'uri_list': ['https://huggingface.co/KaliYuga/PADexpanded/resolve/main/PADexpanded.pt'] },\n",
        "    'pixel_art_diffusion_hard_256': { 'downloaded': False, 'sha': 'be4a9de943ec06eef32c65a1008c60ad017723a4d35dc13169c66bb322234161', 'uri_list': ['https://huggingface.co/KaliYuga/pixel_art_diffusion_hard_256/resolve/main/pixel_art_diffusion_hard_256.pt'] },\n",
        "    'pixel_art_diffusion_soft_256': { 'downloaded': False, 'sha': 'd321590e46b679bf6def1f1914b47c89e762c76f19ab3e3392c8ca07c791039c', 'uri_list': ['https://huggingface.co/KaliYuga/pixel_art_diffusion_soft_256/resolve/main/pixel_art_diffusion_soft_256.pt'] },\n",
        "    'pixelartdiffusion4k': { 'downloaded': False, 'sha': 'a1ba4f13f6dabb72b1064f15d8ae504d98d6192ad343572cc416deda7cccac30', 'uri_list': ['https://huggingface.co/KaliYuga/pixelartdiffusion4k/resolve/main/pixelartdiffusion4k.pt'] },\n",
        "    'watercolordiffusion_2': { 'downloaded': False, 'sha': '49c281b6092c61c49b0f1f8da93af9b94be7e0c20c71e662e2aa26fee0e4b1a9', 'uri_list': ['https://huggingface.co/KaliYuga/watercolordiffusion_2/resolve/main/watercolordiffusion_2.pt'] },\n",
        "    'watercolordiffusion': { 'downloaded': False, 'sha': 'a3e6522f0c8f278f90788298d66383b11ac763dd5e0d62f8252c962c23950bd6', 'uri_list': ['https://huggingface.co/KaliYuga/watercolordiffusion/resolve/main/watercolordiffusion.pt'] },\n",
        "    'PulpSciFiDiffusion': { 'downloaded': False, 'sha': 'b79e62613b9f50b8a3173e5f61f0320c7dbb16efad42a92ec94d014f6e17337f', 'uri_list': ['https://huggingface.co/KaliYuga/PulpSciFiDiffusion/resolve/main/PulpSciFiDiffusion.pt'] },\n",
        "    'secondary': { 'downloaded': False, 'sha': '983e3de6f95c88c81b2ca7ebb2c217933be1973b1ff058776b970f901584613a', 'uri_list': ['https://huggingface.co/spaces/huggi/secondary_model_imagenet_2.pth/resolve/main/secondary_model_imagenet_2.pth', 'https://the-eye.eu/public/AI/models/v-diffusion/secondary_model_imagenet_2.pth', 'https://ipfs.pollinations.ai/ipfs/bafybeibaawhhk7fhyhvmm7x24zwwkeuocuizbqbcg5nqx64jq42j75rdiy/secondary_model_imagenet_2.pth'] },\n",
        "}\n",
        "\n",
        "kaliyuga_pixel_art_model_names = ['pixelartdiffusion_expanded', 'pixel_art_diffusion_hard_256', 'pixel_art_diffusion_soft_256', 'pixelartdiffusion4k', 'PulpSciFiDiffusion']\n",
        "kaliyuga_watercolor_model_names = ['watercolordiffusion', 'watercolordiffusion_2']\n",
        "kaliyuga_pulpscifi_model_names = ['PulpSciFiDiffusion']\n",
        "diffusion_models_256x256_list = ['256x256_diffusion_uncond'] + kaliyuga_pixel_art_model_names + kaliyuga_watercolor_model_names + kaliyuga_pulpscifi_model_names\n",
        "\n",
        "from urllib.parse import urlparse\n",
        "\n",
        "def get_model_filename(diffusion_model_name):\n",
        "    model_uri = diff_model_map[diffusion_model_name]['uri_list'][0]\n",
        "    model_filename = os.path.basename(urlparse(model_uri).path)\n",
        "    return model_filename\n",
        "\n",
        "\n",
        "def download_model(diffusion_model_name, uri_index=0):\n",
        "    if diffusion_model_name != 'custom':\n",
        "        model_filename = get_model_filename(diffusion_model_name)\n",
        "        model_local_path = os.path.join(model_path, model_filename)\n",
        "        if os.path.exists(model_local_path) and check_model_SHA:\n",
        "            print(f'Checking {diffusion_model_name} File')\n",
        "            with open(model_local_path, \"rb\") as f:\n",
        "                bytes = f.read() \n",
        "                hash = hashlib.sha256(bytes).hexdigest()\n",
        "            if hash == diff_model_map[diffusion_model_name]['sha']:\n",
        "                print(f'{diffusion_model_name} SHA matches')\n",
        "                diff_model_map[diffusion_model_name]['downloaded'] = True\n",
        "            else:\n",
        "                print(f\"{diffusion_model_name} SHA doesn't match. Will redownload it.\")\n",
        "        elif os.path.exists(model_local_path) and not check_model_SHA or diff_model_map[diffusion_model_name]['downloaded']:\n",
        "            print(f'{diffusion_model_name} already downloaded. If the file is corrupt, enable check_model_SHA.')\n",
        "            diff_model_map[diffusion_model_name]['downloaded'] = True\n",
        "\n",
        "        if not diff_model_map[diffusion_model_name]['downloaded']:\n",
        "            for model_uri in diff_model_map[diffusion_model_name]['uri_list']:\n",
        "                wget(model_uri, model_path)\n",
        "                if os.path.exists(model_local_path):\n",
        "                    diff_model_map[diffusion_model_name]['downloaded'] = True\n",
        "                    return\n",
        "                else:\n",
        "                    print(f'{diffusion_model_name} model download from {model_uri} failed. Will try any fallback uri.')\n",
        "            print(f'{diffusion_model_name} download failed.')\n",
        "\n",
        "\n",
        "# Download the diffusion model(s)\n",
        "download_model(diffusion_model)\n",
        "if \"True\" in use_secondary_model:\n",
        "    download_model('secondary')\n",
        "\n",
        "model_config = model_and_diffusion_defaults()\n",
        "if diffusion_model == '512x512_diffusion_uncond_finetune_008100':\n",
        "    model_config.update({\n",
        "        'attention_resolutions': '32, 16, 8',\n",
        "        'class_cond': False,\n",
        "        'diffusion_steps': 1000, #No need to edit this, it is taken care of later.\n",
        "        'rescale_timesteps': True,\n",
        "        'timestep_respacing': 250, #No need to edit this, it is taken care of later.\n",
        "        'image_size': 512,\n",
        "        'learn_sigma': True,\n",
        "        'noise_schedule': 'linear',\n",
        "        'num_channels': 256,\n",
        "        'num_head_channels': 64,\n",
        "        'num_res_blocks': 2,\n",
        "        'resblock_updown': True,\n",
        "        'use_checkpoint': use_checkpoint,\n",
        "        'use_fp16': not useCPU,\n",
        "        'use_scale_shift_norm': True,\n",
        "    })\n",
        "elif diffusion_model == '256x256_diffusion_uncond':\n",
        "    model_config.update({\n",
        "        'attention_resolutions': '32, 16, 8',\n",
        "        'class_cond': False,\n",
        "        'diffusion_steps': 1000, #No need to edit this, it is taken care of later.\n",
        "        'rescale_timesteps': True,\n",
        "        'timestep_respacing': 250, #No need to edit this, it is taken care of later.\n",
        "        'image_size': 256,\n",
        "        'learn_sigma': True,\n",
        "        'noise_schedule': 'linear',\n",
        "        'num_channels': 256,\n",
        "        'num_head_channels': 64,\n",
        "        'num_res_blocks': 2,\n",
        "        'resblock_updown': True,\n",
        "        'use_checkpoint': use_checkpoint,\n",
        "        'use_fp16': not useCPU,\n",
        "        'use_scale_shift_norm': True,\n",
        "    })\n",
        "elif diffusion_model == 'portrait_generator_v001':\n",
        "    model_config.update({\n",
        "        'attention_resolutions': '32, 16, 8',\n",
        "        'class_cond': False,\n",
        "        'diffusion_steps': 1000,\n",
        "        'rescale_timesteps': True,\n",
        "        'image_size': 512,\n",
        "        'learn_sigma': True,\n",
        "        'noise_schedule': 'linear',\n",
        "        'num_channels': 128,\n",
        "        'num_heads': 4,\n",
        "        'num_res_blocks': 2,\n",
        "        'resblock_updown': True,\n",
        "        'use_checkpoint': use_checkpoint,\n",
        "        'use_fp16': True,\n",
        "        'use_scale_shift_norm': True,\n",
        "    })\n",
        "else:  # E.g. A model finetuned by KaliYuga\n",
        "    model_config.update({\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,\n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "      })\n",
        "\n",
        "model_default = model_config['image_size']\n",
        "\n",
        "if \"True\" in use_secondary_model:\n",
        "    secondary_model = SecondaryDiffusionImageNet2()\n",
        "    secondary_model.load_state_dict(torch.load(f'{model_path}/secondary_model_imagenet_2.pth', map_location='cpu'))\n",
        "    secondary_model.eval().requires_grad_(False).to(device)\n",
        "\n",
        "clip_models = []\n",
        "if ViTB32: clip_models.append(clip.load('ViT-B/32', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if ViTB16: clip_models.append(clip.load('ViT-B/16', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if ViTL14: clip_models.append(clip.load('ViT-L/14', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if ViTL14_336px: clip_models.append(clip.load('ViT-L/14@336px', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50: clip_models.append(clip.load('RN50', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50x4: clip_models.append(clip.load('RN50x4', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50x16: clip_models.append(clip.load('RN50x16', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN50x64: clip_models.append(clip.load('RN50x64', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "if RN101: clip_models.append(clip.load('RN101', jit=False)[0].eval().requires_grad_(False).to(device))\n",
        "# if ViTB32_laion2b_e16: clip_models.append(open_clip.create_model('ViT-B-32', pretrained='laion2b_e16').eval().requires_grad_(False).to(device))\n",
        "# if ViTB32_laion400m_e31: clip_models.append(open_clip.create_model('ViT-B-32', pretrained='laion400m_e31').eval().requires_grad_(False).to(device))\n",
        "# if ViTB32_laion400m_32: clip_models.append(open_clip.create_model('ViT-B-32', pretrained='laion400m_e32').eval().requires_grad_(False).to(device))\n",
        "# if ViTB32quickgelu_laion400m_e31: clip_models.append(open_clip.create_model('ViT-B-32-quickgelu', pretrained='laion400m_e31').eval().requires_grad_(False).to(device))\n",
        "# if ViTB32quickgelu_laion400m_e32: clip_models.append(open_clip.create_model('ViT-B-32-quickgelu', pretrained='laion400m_e32').eval().requires_grad_(False).to(device))\n",
        "# if ViTB16_laion400m_e31: clip_models.append(open_clip.create_model('ViT-B-16', pretrained='laion400m_e31').eval().requires_grad_(False).to(device))\n",
        "# if ViTB16_laion400m_e32: clip_models.append(open_clip.create_model('ViT-B-16', pretrained='laion400m_e32').eval().requires_grad_(False).to(device))\n",
        "# if RN50_yffcc15m: clip_models.append(open_clip.create_model('RN50', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "# if RN50_cc12m: clip_models.append(open_clip.create_model('RN50', pretrained='cc12m').eval().requires_grad_(False).to(device))\n",
        "# if RN50_quickgelu_yfcc15m: clip_models.append(open_clip.create_model('RN50-quickgelu', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "# if RN50_quickgelu_cc12m: clip_models.append(open_clip.create_model('RN50-quickgelu', pretrained='cc12m').eval().requires_grad_(False).to(device))\n",
        "# if RN101_yfcc15m: clip_models.append(open_clip.create_model('RN101', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "# if RN101_quickgelu_yfcc15m: clip_models.append(open_clip.create_model('RN101-quickgelu', pretrained='yfcc15m').eval().requires_grad_(False).to(device))\n",
        "if VITL14open: clip_models.append(open_clip.create_model('ViT-L-14', pretrained='laion2b_s32b_b79k').eval().requires_grad_(False).to(device))\n",
        "if VITH14open: clip_models.append(open_clip.create_model('ViT-H-14', pretrained='laion2b_s32b_b79k').eval().requires_grad_(False).to(device))\n",
        "if VITG14open: clip_models.append(open_clip.create_model('ViT-g-14', pretrained='laion2b_s12b_b42k').eval().requires_grad_(False).to(device))\n",
        "\n",
        "normalize = T.Normalize(mean=[0.48145466, 0.4578275, 0.40821073], std=[0.26862954, 0.26130258, 0.27577711])\n",
        "lpips_model = lpips.LPIPS(net='vgg').to(device)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CustModelTop"
      },
      "source": [
        "# Custom model settings \n",
        "Modify in accordance with your training settings and run the cell"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CustModel",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Custom Model Settings:**\n",
        "if diffusion_model == 'custom':\n",
        "  model_config.update({\n",
        "          'attention_resolutions': '16',\n",
        "          'class_cond': False,\n",
        "          'diffusion_steps': 1000,\n",
        "          'rescale_timesteps': True,\n",
        "          'timestep_respacing': 'ddim100',\n",
        "          'image_size': 256,\n",
        "          'learn_sigma': True,\n",
        "          'noise_schedule': 'linear',\n",
        "          'num_channels': 128,\n",
        "          'num_heads': 1,\n",
        "          'num_res_blocks': 2,\n",
        "          'use_checkpoint': use_checkpoint,\n",
        "          'use_fp16': True,\n",
        "          'use_scale_shift_norm': False,\n",
        "      })"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SettingsTop"
      },
      "source": [
        "# 3. Settings"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BasicSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Basic Settings:**\n",
        "batch_name = 'MathRock' #@param{type: 'string'}\n",
        "prompt_in_filename = False#@param{type:'boolean'}\n",
        "steps = 250 #@param [25,50,100,150,250,500,1000]{type: 'raw', allow-input: true}\n",
        "width_height = [512,512] #@param{type: 'raw'}\n",
        "clip_guidance_scale = \"[10000]*1000\" #@param{type: 'string'}\n",
        "tv_scale = 59000#@param{type: 'number'}\n",
        "#@markdown setting range_scale to 0 now disables it (as it is not known to currently work), this appears to save a pinch of time\n",
        "range_scale = 0#@param{type: 'number'}\n",
        "sat_scale = 20500#@param{type: 'number'}\n",
        "cutn_batches = \"[3]*1000\"#@param{type: 'string'}\n",
        "skip_augs = False#@param{type: 'boolean'}\n",
        "\n",
        "\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#Get corrected sizes\n",
        "side_x = (width_height[0]//64)*64;\n",
        "side_y = (width_height[1]//64)*64;\n",
        "if side_x != width_height[0] or side_y != width_height[1]:\n",
        "    print(f'Changing output size to {side_x}x{side_y}. Dimensions must by multiples of 64.')\n",
        "\n",
        "#Update Model Settings\n",
        "timestep_respacing = f'ddim{steps}'\n",
        "diffusion_steps = (1000//steps)*steps if steps < 1000 else steps\n",
        "model_config.update({\n",
        "    'timestep_respacing': timestep_respacing,\n",
        "    'diffusion_steps': diffusion_steps,\n",
        "})\n",
        "\n",
        "\n",
        "#Make folder for batch\n",
        "batchFolder = f'{outDirPath}/{batch_name}'\n",
        "createPath(batchFolder)\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "#markdown ###**resize mid run**\n",
        "resize_mid_run = False#param{type:'boolean'}\n",
        "transformation_percent = [0.3]#param\n",
        "newx = 768#param{type:'number'}\n",
        "newy = 960#param{type:'number'}"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AnimSetTop"
      },
      "source": [
        "### Animation Settings"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AnimSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Animation Mode:**\n",
        "animation_mode = 'None' #@param ['None', '2D', '3D', 'Video Input'] {type:'string'}\n",
        "#@markdown *For animation, you probably want to turn `cutn_batches` to 1 to make it quicker.*\n",
        "\n",
        "#@markdown ####**Video Init Basic Settings:**\n",
        "video_init_steps = 100 #@param [25,50,100,150,250,500,1000]{type: 'raw', allow-input: true}\n",
        "video_init_clip_guidance_scale = 1000 #@param{type: 'number'}\n",
        "video_init_tv_scale = 0.1#@param{type: 'number'}\n",
        "video_init_range_scale = 150#@param{type: 'number'}\n",
        "video_init_sat_scale = 300#@param{type: 'number'}\n",
        "video_init_cutn_batches = \"[4]*1000\"#@param{type: 'string'}\n",
        "video_init_skip_steps = 50 #@param{type: 'integer'}\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ####**Video Input Settings:**\n",
        "if is_colab:\n",
        "    video_init_path = \"/content/drive/MyDrive/init.mp4\" #@param {type: 'string'}\n",
        "else:\n",
        "    video_init_path = \"init.mp4\" #@param {type: 'string'}\n",
        "extract_nth_frame = 2 #@param {type: 'number'}\n",
        "persistent_frame_output_in_batch_folder = True #@param {type: 'boolean'}\n",
        "video_init_seed_continuity = False #@param {type: 'boolean'}\n",
        "#@markdown #####**Video Optical Flow Settings:**\n",
        "video_init_flow_warp = True #@param {type: 'boolean'}\n",
        "# Call optical flow from video frames and warp prev frame with flow\n",
        "video_init_flow_blend =  0.999#@param {type: 'number'} #0 - take next frame, 1 - take prev warped frame\n",
        "video_init_check_consistency = False #Insert param here when ready\n",
        "video_init_blend_mode = \"optical flow\" #@param ['None', 'linear', 'optical flow']\n",
        "# Call optical flow from video frames and warp prev frame with flow\n",
        "if animation_mode == \"Video Input\":\n",
        "    if persistent_frame_output_in_batch_folder or (not is_colab): #suggested by Chris the Wizard#8082 at discord\n",
        "        videoFramesFolder = f'{batchFolder}/videoFrames'\n",
        "    else:\n",
        "        videoFramesFolder = f'/content/videoFrames'\n",
        "    createPath(videoFramesFolder)\n",
        "    print(f\"Exporting Video Frames (1 every {extract_nth_frame})...\")\n",
        "    try:\n",
        "        for f in pathlib.Path(f'{videoFramesFolder}').glob('*.jpg'):\n",
        "            f.unlink()\n",
        "    except:\n",
        "        print('')\n",
        "    vf = f'select=not(mod(n\\,{extract_nth_frame}))'\n",
        "    if os.path.exists(video_init_path):\n",
        "        subprocess.run(['ffmpeg', '-i', f'{video_init_path}', '-vf', f'{vf}', '-vsync', 'vfr', '-q:v', '2', '-loglevel', 'error', '-stats', f'{videoFramesFolder}/%04d.jpg'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "    else: \n",
        "        print(f'\\nWARNING!\\n\\nVideo not found: {video_init_path}.\\nPlease check your video path.\\n')\n",
        "    #!ffmpeg -i {video_init_path} -vf {vf} -vsync vfr -q:v 2 -loglevel error -stats {videoFramesFolder}/%04d.jpg\n",
        "\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ####**2D Animation Settings:**\n",
        "#@markdown `zoom` is a multiplier of dimensions, 1 is no zoom.\n",
        "#@markdown All rotations are provided in degrees.\n",
        "\n",
        "key_frames = True #@param {type:\"boolean\"}\n",
        "max_frames = 10000#@param {type:\"number\"}\n",
        "\n",
        "if animation_mode == \"Video Input\":\n",
        "    max_frames = len(glob(f'{videoFramesFolder}/*.jpg'))\n",
        "\n",
        "interp_spline = 'Linear' #Do not change, currently will not look good. param ['Linear','Quadratic','Cubic']{type:\"string\"}\n",
        "angle = \"0:(0)\"#@param {type:\"string\"}\n",
        "zoom = \"0: (1), 10: (1.05)\"#@param {type:\"string\"}\n",
        "translation_x = \"0: (0)\"#@param {type:\"string\"}\n",
        "translation_y = \"0: (0)\"#@param {type:\"string\"}\n",
        "translation_z = \"0: (10.0)\"#@param {type:\"string\"}\n",
        "rotation_3d_x = \"0: (0)\"#@param {type:\"string\"}\n",
        "rotation_3d_y = \"0: (0)\"#@param {type:\"string\"}\n",
        "rotation_3d_z = \"0: (0)\"#@param {type:\"string\"}\n",
        "midas_depth_model = \"dpt_large\"#@param {type:\"string\"}\n",
        "midas_weight = 0.3#@param {type:\"number\"}\n",
        "near_plane = 200#@param {type:\"number\"}\n",
        "far_plane = 10000#@param {type:\"number\"}\n",
        "fov = 40#@param {type:\"number\"}\n",
        "padding_mode = 'border'#@param {type:\"string\"}\n",
        "sampling_mode = 'bicubic'#@param {type:\"string\"}\n",
        "\n",
        "#======= TURBO MODE\n",
        "#@markdown ---\n",
        "#@markdown ####**Turbo Mode (3D anim only):**\n",
        "#@markdown (Starts after frame 10,) skips diffusion steps and just uses depth map to warp images for skipped frames.\n",
        "#@markdown Speeds up rendering by 2x-4x, and may improve image coherence between frames.\n",
        "#@markdown For different settings tuned for Turbo Mode, refer to the original Disco-Turbo Github: https://github.com/zippy731/disco-diffusion-turbo\n",
        "\n",
        "turbo_mode = False #@param {type:\"boolean\"}\n",
        "turbo_steps = \"3\" #@param [\"2\",\"3\",\"4\",\"5\",\"6\"] {type:\"string\"}\n",
        "turbo_preroll = 10 # frames\n",
        "\n",
        "#insist turbo be used only w 3d anim.\n",
        "if turbo_mode and animation_mode != '3D':\n",
        "    print('=====')\n",
        "    print('Turbo mode only available with 3D animations. Disabling Turbo.')\n",
        "    print('=====')\n",
        "    turbo_mode = False\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ####**Coherency Settings:**\n",
        "#@markdown `frame_scale` tries to guide the new frame to looking like the old one. A good default is 1500.\n",
        "frames_scale = 1500 #@param{type: 'integer'}\n",
        "#@markdown `frame_skip_steps` will blur the previous frame - higher values will flicker less but struggle to add enough new detail to zoom into.\n",
        "frames_skip_steps = '60%' #@param ['40%', '50%', '60%', '70%', '80%'] {type: 'string'}\n",
        "\n",
        "#@markdown ####**Video Init Coherency Settings:**\n",
        "#@markdown `frame_scale` tries to guide the new frame to looking like the old one. A good default is 1500.\n",
        "video_init_frames_scale = 15000 #@param{type: 'integer'}\n",
        "#@markdown `frame_skip_steps` will blur the previous frame - higher values will flicker less but struggle to add enough new detail to zoom into.\n",
        "video_init_frames_skip_steps = '70%' #@param ['40%', '50%', '60%', '70%', '80%'] {type: 'string'}\n",
        "\n",
        "\n",
        "#============= NOISE INJECTOR\n",
        "\n",
        "#@markdown ---\n",
        "#@markdown ####**Noise Injector (3D anim only):**\n",
        "#@markdown (experimental!) Adds noise if dead spots detected\n",
        "noise_injector = False #@param {type:\"boolean\"}\n",
        "noise_injector_threshold =  35#@param{type: 'integer'}\n",
        "#@markdown `noise_injector_threshold` if edge% falls below this %, will boost next frame. 20 is a good starting value.\n",
        "noise_injector_mix = 0.75 #@param{type: 'number'}\n",
        "#@markdown `noise_injector_mix` overall wet/dry mix for noise effect 0 - 1.0 \n",
        "noise_injector_px_size = 4 #@param{type: 'integer'}\n",
        "#@markdown `noise_injector_px_size` size of noise color blocks.\n",
        "noise_injector_spice = .10 #@param{type: 'number'}\n",
        "#@markdown `noise_injector_spice` adds random noise to color-matched noise. 0 - 1.0 ratio.\n",
        "noise_injector_mask_feather = 10 #@param{type: 'integer'}\n",
        "#@markdown `noise_injector_mask_feather` introduce soft edges at noise mask distance in px. \n",
        "noise_injector_sigma = 32 #@param{type: 'integer'}\n",
        "#@markdown `noise_injector_sigma` = size of 'clumps' in noise. higher is larger, 16-32 are good starting pt \n",
        "\n",
        "#@markdown ####**Copy palette conform animation frames to color mix from palette_image (2D/3D/video only):**\n",
        "copy_palette = False#@param{type: 'boolean'}\n",
        "copy_palette_image = \"/content/drive/MyDrive/AI/Disco_Diffusion/init_images/beach.jpg\" #@param{type: 'string'}\n",
        "\n",
        "#======= VR MODE\n",
        "#@markdown ---\n",
        "#@markdown ####**VR Mode (3D anim only):**\n",
        "#@markdown Enables stereo rendering of left/right eye views (supporting Turbo) which use a different (fish-eye) camera projection matrix.   \n",
        "#@markdown Note the images you're prompting will work better if they have some inherent wide-angle aspect\n",
        "#@markdown The generated images will need to be combined into left/right videos. These can then be stitched into the VR180 format.\n",
        "#@markdown Google made the VR180 Creator tool but subsequently stopped supporting it. It's available for download in a few places including https://www.patrickgrunwald.de/vr180-creator-download\n",
        "#@markdown The tool is not only good for stitching (videos and photos) but also for adding the correct metadata into existing videos, which is needed for services like YouTube to identify the format correctly.\n",
        "#@markdown Watching YouTube VR videos isn't necessarily the easiest depending on your headset. For instance Oculus have a dedicated media studio and store which makes the files easier to access on a Quest https://creator.oculus.com/manage/mediastudio/\n",
        "#@markdown \n",
        "#@markdown The command to get ffmpeg to concat your frames for each eye is in the form: `ffmpeg -framerate 15 -i frame_%4d_l.png l.mp4` (repeat for r)\n",
        "\n",
        "vr_mode = False #@param {type:\"boolean\"}\n",
        "#@markdown `vr_eye_angle` is the y-axis rotation of the eyes towards the center\n",
        "vr_eye_angle = 0.5 #@param{type:\"number\"}\n",
        "#@markdown interpupillary distance (between the eyes)\n",
        "vr_ipd = 5.0 #@param{type:\"number\"}\n",
        "\n",
        "#insist VR be used only w 3d anim.\n",
        "if vr_mode and animation_mode != '3D':\n",
        "    print('=====')\n",
        "    print('VR mode only available with 3D animations. Disabling VR.')\n",
        "    print('=====')\n",
        "    vr_mode = False\n",
        "\n",
        "\n",
        "def parse_key_frames(string, prompt_parser=None):\n",
        "    \"\"\"Given a string representing frame numbers paired with parameter values at that frame,\n",
        "    return a dictionary with the frame numbers as keys and the parameter values as the values.\n",
        "\n",
        "    Parameters\n",
        "    ----------\n",
        "    string: string\n",
        "        Frame numbers paired with parameter values at that frame number, in the format\n",
        "        'framenumber1: (parametervalues1), framenumber2: (parametervalues2), ...'\n",
        "    prompt_parser: function or None, optional\n",
        "        If provided, prompt_parser will be applied to each string of parameter values.\n",
        "    \n",
        "    Returns\n",
        "    -------\n",
        "    dict\n",
        "        Frame numbers as keys, parameter values at that frame number as values\n",
        "\n",
        "    Raises\n",
        "    ------\n",
        "    RuntimeError\n",
        "        If the input string does not match the expected format.\n",
        "    \n",
        "    Examples\n",
        "    --------\n",
        "    >>> parse_key_frames(\"10:(Apple: 1| Orange: 0), 20: (Apple: 0| Orange: 1| Peach: 1)\")\n",
        "    {10: 'Apple: 1| Orange: 0', 20: 'Apple: 0| Orange: 1| Peach: 1'}\n",
        "\n",
        "    >>> parse_key_frames(\"10:(Apple: 1| Orange: 0), 20: (Apple: 0| Orange: 1| Peach: 1)\", prompt_parser=lambda x: x.lower()))\n",
        "    {10: 'apple: 1| orange: 0', 20: 'apple: 0| orange: 1| peach: 1'}\n",
        "    \"\"\"\n",
        "    import re\n",
        "    pattern = r'((?P<frame>[0-9]+):[\\s]*[\\(](?P<param>[\\S\\s]*?)[\\)])'\n",
        "    frames = dict()\n",
        "    for match_object in re.finditer(pattern, string):\n",
        "        frame = int(match_object.groupdict()['frame'])\n",
        "        param = match_object.groupdict()['param']\n",
        "        if prompt_parser:\n",
        "            frames[frame] = prompt_parser(param)\n",
        "        else:\n",
        "            frames[frame] = param\n",
        "\n",
        "    if frames == {} and len(string) != 0:\n",
        "        raise RuntimeError('Key Frame string not correctly formatted')\n",
        "    return frames\n",
        "\n",
        "def get_inbetweens(key_frames, integer=False):\n",
        "    \"\"\"Given a dict with frame numbers as keys and a parameter value as values,\n",
        "    return a pandas Series containing the value of the parameter at every frame from 0 to max_frames.\n",
        "    Any values not provided in the input dict are calculated by linear interpolation between\n",
        "    the values of the previous and next provided frames. If there is no previous provided frame, then\n",
        "    the value is equal to the value of the next provided frame, or if there is no next provided frame,\n",
        "    then the value is equal to the value of the previous provided frame. If no frames are provided,\n",
        "    all frame values are NaN.\n",
        "\n",
        "    Parameters\n",
        "    ----------\n",
        "    key_frames: dict\n",
        "        A dict with integer frame numbers as keys and numerical values of a particular parameter as values.\n",
        "    integer: Bool, optional\n",
        "        If True, the values of the output series are converted to integers.\n",
        "        Otherwise, the values are floats.\n",
        "    \n",
        "    Returns\n",
        "    -------\n",
        "    pd.Series\n",
        "        A Series with length max_frames representing the parameter values for each frame.\n",
        "    \n",
        "    Examples\n",
        "    --------\n",
        "    >>> max_frames = 5\n",
        "    >>> get_inbetweens({1: 5, 3: 6})\n",
        "    0    5.0\n",
        "    1    5.0\n",
        "    2    5.5\n",
        "    3    6.0\n",
        "    4    6.0\n",
        "    dtype: float64\n",
        "\n",
        "    >>> get_inbetweens({1: 5, 3: 6}, integer=True)\n",
        "    0    5\n",
        "    1    5\n",
        "    2    5\n",
        "    3    6\n",
        "    4    6\n",
        "    dtype: int64\n",
        "    \"\"\"\n",
        "    key_frame_series = pd.Series([np.nan for a in range(max_frames)])\n",
        "\n",
        "    for i, value in key_frames.items():\n",
        "        key_frame_series[i] = value\n",
        "    key_frame_series = key_frame_series.astype(float)\n",
        "    \n",
        "    interp_method = interp_spline\n",
        "\n",
        "    if interp_method == 'Cubic' and len(key_frames.items()) <=3:\n",
        "      interp_method = 'Quadratic'\n",
        "    \n",
        "    if interp_method == 'Quadratic' and len(key_frames.items()) <= 2:\n",
        "      interp_method = 'Linear'\n",
        "      \n",
        "    \n",
        "    key_frame_series[0] = key_frame_series[key_frame_series.first_valid_index()]\n",
        "    key_frame_series[max_frames-1] = key_frame_series[key_frame_series.last_valid_index()]\n",
        "    # key_frame_series = key_frame_series.interpolate(method=intrp_method,order=1, limit_direction='both')\n",
        "    key_frame_series = key_frame_series.interpolate(method=interp_method.lower(),limit_direction='both')\n",
        "    if integer:\n",
        "        return key_frame_series.astype(int)\n",
        "    return key_frame_series\n",
        "\n",
        "def split_prompts(prompts):\n",
        "    prompt_series = pd.Series([np.nan for a in range(max_frames)])\n",
        "    for i, prompt in prompts.items():\n",
        "        prompt_series[i] = prompt\n",
        "    # prompt_series = prompt_series.astype(str)\n",
        "    prompt_series = prompt_series.ffill().bfill()\n",
        "    return prompt_series\n",
        "\n",
        "if key_frames:\n",
        "    try:\n",
        "        angle_series = get_inbetweens(parse_key_frames(angle))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `angle` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `angle` as \"\n",
        "            f'\"0: ({angle})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        angle = f\"0: ({angle})\"\n",
        "        angle_series = get_inbetweens(parse_key_frames(angle))\n",
        "\n",
        "    try:\n",
        "        zoom_series = get_inbetweens(parse_key_frames(zoom))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `zoom` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `zoom` as \"\n",
        "            f'\"0: ({zoom})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        zoom = f\"0: ({zoom})\"\n",
        "        zoom_series = get_inbetweens(parse_key_frames(zoom))\n",
        "\n",
        "    try:\n",
        "        translation_x_series = get_inbetweens(parse_key_frames(translation_x))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `translation_x` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `translation_x` as \"\n",
        "            f'\"0: ({translation_x})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        translation_x = f\"0: ({translation_x})\"\n",
        "        translation_x_series = get_inbetweens(parse_key_frames(translation_x))\n",
        "\n",
        "    try:\n",
        "        translation_y_series = get_inbetweens(parse_key_frames(translation_y))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `translation_y` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `translation_y` as \"\n",
        "            f'\"0: ({translation_y})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        translation_y = f\"0: ({translation_y})\"\n",
        "        translation_y_series = get_inbetweens(parse_key_frames(translation_y))\n",
        "\n",
        "    try:\n",
        "        translation_z_series = get_inbetweens(parse_key_frames(translation_z))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `translation_z` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `translation_z` as \"\n",
        "            f'\"0: ({translation_z})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        translation_z = f\"0: ({translation_z})\"\n",
        "        translation_z_series = get_inbetweens(parse_key_frames(translation_z))\n",
        "\n",
        "    try:\n",
        "        rotation_3d_x_series = get_inbetweens(parse_key_frames(rotation_3d_x))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `rotation_3d_x` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `rotation_3d_x` as \"\n",
        "            f'\"0: ({rotation_3d_x})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        rotation_3d_x = f\"0: ({rotation_3d_x})\"\n",
        "        rotation_3d_x_series = get_inbetweens(parse_key_frames(rotation_3d_x))\n",
        "\n",
        "    try:\n",
        "        rotation_3d_y_series = get_inbetweens(parse_key_frames(rotation_3d_y))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `rotation_3d_y` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `rotation_3d_y` as \"\n",
        "            f'\"0: ({rotation_3d_y})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        rotation_3d_y = f\"0: ({rotation_3d_y})\"\n",
        "        rotation_3d_y_series = get_inbetweens(parse_key_frames(rotation_3d_y))\n",
        "\n",
        "    try:\n",
        "        rotation_3d_z_series = get_inbetweens(parse_key_frames(rotation_3d_z))\n",
        "    except RuntimeError as e:\n",
        "        print(\n",
        "            \"WARNING: You have selected to use key frames, but you have not \"\n",
        "            \"formatted `rotation_3d_z` correctly for key frames.\\n\"\n",
        "            \"Attempting to interpret `rotation_3d_z` as \"\n",
        "            f'\"0: ({rotation_3d_z})\"\\n'\n",
        "            \"Please read the instructions to find out how to use key frames \"\n",
        "            \"correctly.\\n\"\n",
        "        )\n",
        "        rotation_3d_z = f\"0: ({rotation_3d_z})\"\n",
        "        rotation_3d_z_series = get_inbetweens(parse_key_frames(rotation_3d_z))\n",
        "\n",
        "else:\n",
        "    angle = float(angle)\n",
        "    zoom = float(zoom)\n",
        "    translation_x = float(translation_x)\n",
        "    translation_y = float(translation_y)\n",
        "    translation_z = float(translation_z)\n",
        "    rotation_3d_x = float(rotation_3d_x)\n",
        "    rotation_3d_y = float(rotation_3d_y)\n",
        "    rotation_3d_z = float(rotation_3d_z)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "InstallRAFT",
        "cellView": "form"
      },
      "source": [
        "#@title Install RAFT for Video input animation mode only\n",
        "#@markdown Run once per session. Doesn't download again if model path exists.\n",
        "#@markdown Use force download to reload raft models if needed\n",
        "force_download = False #@param {type:'boolean'}\n",
        "if animation_mode == 'Video Input':\n",
        "    try:\n",
        "        from raft import RAFT\n",
        "    except:\n",
        "        if not os.path.exists(os.path.join(PROJECT_DIR, 'RAFT')):\n",
        "            gitclone('https://github.com/princeton-vl/RAFT', os.path.join(PROJECT_DIR, 'RAFT'))\n",
        "        sys.path.append(f'{PROJECT_DIR}/RAFT')\n",
        "\n",
        "    if (not (os.path.exists(f'{root_path}/RAFT/models'))) or force_download:\n",
        "        createPath(f'{root_path}/RAFT')\n",
        "        os.chdir(f'{root_path}/RAFT')\n",
        "        sub_p_res = subprocess.run(['bash', f'{PROJECT_DIR}/RAFT/download_models.sh'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "        print(sub_p_res)\n",
        "        os.chdir(PROJECT_DIR)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FlowFns1",
        "cellView": "form"
      },
      "source": [
        "#@title Define optical flow functions for Video input animation mode only\n",
        "if animation_mode == 'Video Input':\n",
        "    in_path = videoFramesFolder\n",
        "    flo_folder = f'{in_path}/out_flo_fwd'\n",
        "    path = f'{PROJECT_DIR}/RAFT/core'\n",
        "    import sys\n",
        "    sys.path.append(f'{PROJECT_DIR}/RAFT/core')\n",
        "    os.chdir(f'{PROJECT_DIR}/RAFT/core')\n",
        "    print(os.getcwd())\n",
        "  \n",
        "    print(\"Renaming RAFT core's utils.utils to raftutils.utils (to avoid a naming conflict with AdaBins)\")\n",
        "    if not os.path.exists(f'{PROJECT_DIR}/RAFT/core/raftutils'):\n",
        "        os.rename(f'{PROJECT_DIR}/RAFT/core/utils', f'{PROJECT_DIR}/RAFT/core/raftutils')\n",
        "        sub_p_res = subprocess.run(['sed', '-i', 's/from utils.utils/from raftutils.utils/g', f'{PROJECT_DIR}/RAFT/core/corr.py'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "        sub_p_res = subprocess.run(['sed', '-i', 's/from utils.utils/from raftutils.utils/g', f'{PROJECT_DIR}/RAFT/core/raft.py'], stdout=subprocess.PIPE).stdout.decode('utf-8')\n",
        "\n",
        "    from raftutils.utils import InputPadder\n",
        "    from raft import RAFT\n",
        "    from raftutils import flow_viz\n",
        "    import numpy as np\n",
        "    import argparse, PIL, cv2\n",
        "    from PIL import Image\n",
        "    from tqdm.notebook import tqdm\n",
        "    from glob import glob\n",
        "    import torch\n",
        "  \n",
        "    args2 = argparse.Namespace()\n",
        "    args2.small = False\n",
        "    args2.mixed_precision = True\n",
        "  \n",
        "  \n",
        "    TAG_CHAR = np.array([202021.25], np.float32)\n",
        "  \n",
        "    def writeFlow(filename,uv,v=None):\n",
        "        \"\"\" \n",
        "        https://github.com/NVIDIA/flownet2-pytorch/blob/master/utils/flow_utils.py\n",
        "        Copyright 2017 NVIDIA CORPORATION\n",
        "  \n",
        "        Licensed under the Apache License, Version 2.0 (the \"License\");\n",
        "        you may not use this file except in compliance with the License.\n",
        "        You may obtain a copy of the License at\n",
        "  \n",
        "            http://www.apache.org/licenses/LICENSE-2.0\n",
        "  \n",
        "        Unless required by applicable law or agreed to in writing, software\n",
        "        distributed under the License is distributed on an \"AS IS\" BASIS,\n",
        "        WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
        "        See the License for the specific language governing permissions and\n",
        "        limitations under the License.\n",
        "        \n",
        "        Write optical flow to file.\n",
        "        \n",
        "        If v is None, uv is assumed to contain both u and v channels,\n",
        "        stacked in depth.\n",
        "        Original code by Deqing Sun, adapted from Daniel Scharstein.\n",
        "        \"\"\"\n",
        "        nBands = 2\n",
        "  \n",
        "        if v is None:\n",
        "            assert(uv.ndim == 3)\n",
        "            assert(uv.shape[2] == 2)\n",
        "            u = uv[:,:,0]\n",
        "            v = uv[:,:,1]\n",
        "        else:\n",
        "            u = uv\n",
        "  \n",
        "        assert(u.shape == v.shape)\n",
        "        height,width = u.shape\n",
        "        f = open(filename,'wb')\n",
        "        # write the header\n",
        "        f.write(TAG_CHAR)\n",
        "        np.array(width).astype(np.int32).tofile(f)\n",
        "        np.array(height).astype(np.int32).tofile(f)\n",
        "        # arrange into matrix form\n",
        "        tmp = np.zeros((height, width*nBands))\n",
        "        tmp[:,np.arange(width)*2] = u\n",
        "        tmp[:,np.arange(width)*2 + 1] = v\n",
        "        tmp.astype(np.float32).tofile(f)\n",
        "        f.close()\n",
        "  \n",
        "    def load_img(img, size):\n",
        "        img = Image.open(img).convert('RGB').resize(size)\n",
        "        return torch.from_numpy(np.array(img)).permute(2,0,1).float()[None,...].cuda()\n",
        "  \n",
        "    def get_flow(frame1, frame2, model, iters=20):\n",
        "        padder = InputPadder(frame1.shape)\n",
        "        frame1, frame2 = padder.pad(frame1, frame2)\n",
        "        _, flow12 = model(frame1, frame2, iters=iters, test_mode=True)\n",
        "        flow12 = flow12[0].permute(1, 2, 0).detach().cpu().numpy()\n",
        "  \n",
        "        return flow12\n",
        "  \n",
        "    def warp_flow(img, flow):\n",
        "        h, w = flow.shape[:2]\n",
        "        flow = flow.copy()\n",
        "        flow[:, :, 0] += np.arange(w)\n",
        "        flow[:, :, 1] += np.arange(h)[:, np.newaxis]\n",
        "        res = cv2.remap(img, flow, None, cv2.INTER_LINEAR)\n",
        "        return res\n",
        "  \n",
        "    def makeEven(_x):\n",
        "        return _x if (_x % 2 == 0) else _x+1\n",
        "  \n",
        "    def fit(img,maxsize=512):\n",
        "        maxdim = max(*img.size)\n",
        "        if maxdim>maxsize:\n",
        "            # if True:\n",
        "            ratio = maxsize/maxdim\n",
        "            x,y = img.size\n",
        "            size = (makeEven(int(x*ratio)),makeEven(int(y*ratio))) \n",
        "            img = img.resize(size)\n",
        "        return img\n",
        "  \n",
        "    def warp(frame1, frame2, flo_path, blend=0.5, weights_path=None):\n",
        "        flow21 = np.load(flo_path)\n",
        "        frame1pil = np.array(frame1.convert('RGB').resize((flow21.shape[1],flow21.shape[0])))\n",
        "        frame1_warped21 = warp_flow(frame1pil, flow21)\n",
        "        # frame2pil = frame1pil\n",
        "        frame2pil = np.array(frame2.convert('RGB').resize((flow21.shape[1],flow21.shape[0])))\n",
        "    \n",
        "        if weights_path:\n",
        "            # TBD\n",
        "            pass\n",
        "        else:\n",
        "            blended_w = frame2pil*(1-blend) + frame1_warped21*(blend)\n",
        "  \n",
        "        return  PIL.Image.fromarray(blended_w.astype('uint8'))\n",
        "  \n",
        "    in_path = videoFramesFolder\n",
        "    flo_folder = f'{in_path}/out_flo_fwd'\n",
        "  \n",
        "    temp_flo = in_path+'/temp_flo'\n",
        "    flo_fwd_folder = in_path+'/out_flo_fwd'\n",
        "    # TBD flow backwards!\n",
        "  \n",
        "    os.chdir(PROJECT_DIR)"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FlowFns2",
        "cellView": "form"
      },
      "source": [
        "#@title Generate optical flow and consistency maps\n",
        "#@markdown Run once per init video\n",
        "\n",
        "if animation_mode == \"Video Input\":\n",
        "    import gc\n",
        "\n",
        "    force_flow_generation = False #@param {type:'boolean'}\n",
        "    in_path = videoFramesFolder\n",
        "    flo_folder = f'{in_path}/out_flo_fwd'\n",
        "\n",
        "    if not video_init_flow_warp:\n",
        "        print('video_init_flow_warp not set, skipping')\n",
        "\n",
        "    if (animation_mode == 'Video Input') and (video_init_flow_warp):\n",
        "        flows = glob(flo_folder+'/*.*')\n",
        "        if (len(flows)>0) and not force_flow_generation:\n",
        "            print(f'Skipping flow generation:\\nFound {len(flows)} existing flow files in current working folder: {flo_folder}.\\nIf you wish to generate new flow files, check force_flow_generation and run this cell again.')\n",
        "    \n",
        "        if (len(flows)==0) or force_flow_generation:\n",
        "            frames = sorted(glob(in_path+'/*.*'));\n",
        "            if len(frames)<2: \n",
        "                print(f'WARNING!\\nCannot create flow maps: Found {len(frames)} frames extracted from your video input.\\nPlease check your video path.')\n",
        "            if len(frames)>=2:\n",
        "        \n",
        "                raft_model = torch.nn.DataParallel(RAFT(args2))\n",
        "                raft_model.load_state_dict(torch.load(f'{root_path}/RAFT/models/raft-things.pth'))\n",
        "                raft_model = raft_model.module.cuda().eval()\n",
        "        \n",
        "                for f in pathlib.Path(f'{flo_fwd_folder}').glob('*.*'):\n",
        "                    f.unlink()\n",
        "        \n",
        "                temp_flo = in_path+'/temp_flo'\n",
        "                flo_fwd_folder = in_path+'/out_flo_fwd'\n",
        "        \n",
        "                createPath(flo_fwd_folder)\n",
        "                createPath(temp_flo)\n",
        "        \n",
        "                # TBD Call out to a consistency checker?\n",
        "        \n",
        "                framecount = 0\n",
        "                for frame1, frame2 in tqdm(zip(frames[:-1], frames[1:]), total=len(frames)-1):\n",
        "        \n",
        "                    out_flow21_fn = f\"{flo_fwd_folder}/{frame1.split('/')[-1]}\"\n",
        "            \n",
        "                    frame1 = load_img(frame1, width_height)\n",
        "                    frame2 = load_img(frame2, width_height)\n",
        "            \n",
        "                    flow21 = get_flow(frame2, frame1, raft_model)\n",
        "                    np.save(out_flow21_fn, flow21)\n",
        "            \n",
        "                    if video_init_check_consistency:\n",
        "                        # TBD\n",
        "                        pass\n",
        "\n",
        "                del raft_model \n",
        "                gc.collect()"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ExtraSetTop"
      },
      "source": [
        "### Extra Settings\n",
        " Partial Saves, Advanced Settings, Cutn Scheduling"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ExtraSettings",
        "cellView": "form"
      },
      "source": [
        "#@markdown ####**Saving:**\n",
        "\n",
        "intermediate_saves = 0#@param{type: 'raw'}\n",
        "intermediates_in_subfolder = True #@param{type: 'boolean'}\n",
        "#@markdown Intermediate steps will save a copy at your specified intervals. You can either format it as a single integer or a list of specific steps \n",
        "\n",
        "#@markdown A value of `2` will save a copy at 33% and 66%. 0 will save none.\n",
        "\n",
        "#@markdown A value of `[5, 9, 34, 45]` will save at steps 5, 9, 34, and 45. (Make sure to include the brackets)\n",
        "\n",
        "\n",
        "if type(intermediate_saves) is not list:\n",
        "    if intermediate_saves:\n",
        "        steps_per_checkpoint = math.floor((steps - skip_steps - 1) // (intermediate_saves+1))\n",
        "        steps_per_checkpoint = steps_per_checkpoint if steps_per_checkpoint > 0 else 1\n",
        "        print(f'Will save every {steps_per_checkpoint} steps')\n",
        "    else:\n",
        "        steps_per_checkpoint = steps+10\n",
        "else:\n",
        "    steps_per_checkpoint = None\n",
        "\n",
        "if intermediate_saves and intermediates_in_subfolder is True:\n",
        "    partialFolder = f'{batchFolder}/partials'\n",
        "    createPath(partialFolder)\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#@markdown ####**Advanced Settings:**\n",
        "#@markdown *There are a few extra advanced settings available if you double click this cell.*\n",
        "\n",
        "perlin_init = True  #@param{type: 'boolean'}\n",
        "perlin_mode = 'mixed' #@param ['mixed', 'color', 'gray']\n",
        "set_seed = 'random_seed' #@param{type: 'string'}\n",
        "eta = 0.75 #@param{type: 'number'}\n",
        "clamp_grad = True #@param{type: 'boolean'}\n",
        "clamp_max = \"[0.18]*1000\" #@param{type: 'string'}\n",
        "\n",
        "\n",
        "\n",
        "### EXTRA ADVANCED SETTINGS:\n",
        "randomize_class = True\n",
        "clip_denoised = False#@param{type: 'boolean'}\n",
        "fuzzy_prompt = False\n",
        "rand_mag = 0.05\n",
        "\n",
        "\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#markdown ####**Transformation Settings:**\n",
        "use_vertical_symmetry = False #param {type:\"boolean\"}\n",
        "use_horizontal_symmetry = False #param {type:\"boolean\"}\n",
        "#transformation_percent = [0.09] #param\n",
        "\n",
        "\n",
        "#@markdown ####**Symmetry Settings:**\n",
        "#@markdown *Enable symmetry_loss to mirror left/right side of image until step symm_switch.*\n",
        "\n",
        "symmetry_loss = False#@param{type: 'boolean'}\n",
        "symm_loss_scale =  2000#@param\n",
        "symm_switch = 86#@param\n",
        "symm_switch = 100.*(1. - (symm_switch/steps)) \n",
        "if symmetry_loss:\n",
        "    print(f\"Symmetry ends at {100-symm_switch}%\")\n",
        "\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "#markdown ####**Cutn Scheduling:**\n",
        "#markdown Format: `[40]*400+[20]*600` = 40 cuts for the first 400 /1000 steps, then 20 for the last 600/1000\n",
        "\n",
        "#markdown cut_overview and cut_innercut are cumulative for total cutn on any given step. Overview cuts see the entire image and are good for early structure, innercuts are your standard cutn.\n",
        "\n",
        "\n"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "source": [
        "###Cut Settings"
      ],
      "metadata": {
        "id": "HFKimcseKtCN"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "\n",
        "\n",
        "weighting = 'manual'#param['regular','manual','adaptive']\n",
        "\n",
        "\n",
        "cut_overview = \"[6]*200+[6]*200+[4]*300+[2]*300\"  #@param {type: 'string'}       \n",
        "cut_innercut = \"[10]*200+[10]*200+[12]*300+[14]*300\" #@param {type: 'string'} \n",
        "\n",
        "\n",
        "\n",
        "overview_weight = \"[16]*200+[7]*100+[6]*100+[1.5]*100+[1]*500\"#@param{type: 'string'}\n",
        "\n",
        "\n",
        "#@markdown this setting will set a portion of overview cuts to be executed as max_size cuts instead, scheduled to favor max_size as the run progresses\n",
        "percent_overview = \"[0.8]*190+[0.8]*70+[0.7]*140+[0.5]*600\" #@param{type: 'string'}\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        " \n",
        "\n",
        "\n",
        "\n",
        "\n",
        "#@markdown several ways of distributing cut sizes\n",
        "\n",
        "\n",
        "cut_ic_pow = \"[3.5]*250+[4.5]*160+[20]*590\"#@param {type: 'string'}  \n",
        "\n",
        "#@markdown floor size sets a minimum size for cuts to be executed at.\n",
        "floor_size = \"[0]*1000\"#@param {type: 'string'}\n",
        "#floor_size = \"[0.1]*150+[0.1]*150+[0.1]*100+[0.05]*100+[0]*500\"#param {type: 'string'}\n",
        "\n",
        "#@markdown ceiling is a maximum size for cuts\n",
        "ceiling = 1.0#@param{type:'number'}\n",
        "\n",
        "\n",
        "autofloor = False#@param{type:\"boolean\"}\n",
        "\n",
        "basefloor = (((0.1225 * side_x * side_y)-224)/((side_x * side_y)-224))*0.77\n",
        "\n",
        "tempfloor = f'[{basefloor}]*400+[{basefloor/2}]*100+[0]*500'\n",
        "\n",
        "if autofloor:\n",
        "  if (((side_x * side_y)**0.5) > 640):\n",
        "    floor_size = tempfloor\n",
        "  else:\n",
        "    floor_size = \"[0]*1000\"\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "ic_gray_p = \"[0.2]*400+[0]*600\"#@param {type: 'string'}\n",
        "ov_gray_p = \"[0.2]*100+[0.1]*300+[0]*600\"#@param {type: 'string'}\n",
        "\n",
        "\n",
        "\n",
        "cut_weight_multiplier=\"[0.95]*200+[0.55]*250+[0.0]*550\"\n",
        "\n",
        "\n",
        "overview_type = \"jaxy2\"#@param['variable zoom','jaxy','jaxy2','regular']\n",
        "if overview_type == \"variable zoom\":\n",
        "  zoom_out_overviews = True\n",
        "  jaxy_overviews = False\n",
        "  jaxy_overviews2 = False\n",
        "elif overview_type == \"jaxy\":\n",
        "  zoom_out_overviews = False\n",
        "  jaxy_overviews = True\n",
        "  jaxy_overviews2 = False\n",
        "elif overview_type == \"jaxy2\":\n",
        "  zoom_out_overviews = False\n",
        "  jaxy_overviews = False\n",
        "  jaxy_overviews2 = True\n",
        "else:\n",
        "  zoom_out_overviews = False\n",
        "  jaxy_overviews = False\n",
        "  jaxy_overviews2 = False\n",
        "\n",
        "hflip = True#param{type:'boolean'}\n",
        "\n",
        "#@markdown affine slightly rotates overview cuts, this seems to be useful, but also the most time-consuming augmentation\n",
        "affine = True#@param{type:'boolean'}\n",
        "\n",
        "\n",
        "#markdown normalize loss. generally want to normalize, but might be good to inflate the loss early? may need to reduce cgs to compensate\n",
        "normalizeloss = True#param{type:'boolean'}\n",
        "\n",
        "col_jitter = 0.08\n",
        "pixel_jitter = 0.027\n",
        "degree_tilt = 10\n",
        "\n",
        "percent_stop_affine = 50\n",
        "\n",
        "perspective = False\n",
        "\n",
        "\n",
        "if weighting == 'adaptive':\n",
        "  adaptive_weight = True\n",
        "  manual_weight = False\n",
        "elif weighting == 'manual':\n",
        "  adaptive_weight = False\n",
        "  manual_weight = True\n",
        "else:\n",
        "  adaptive_weight = False\n",
        "  manual_weight = False\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "9snNv9k7KYVI",
        "cellView": "form"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "###bucket cuts"
      ],
      "metadata": {
        "id": "wTJTMhofQNQW"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@markdown leave this at none to disable, otherwise will use set distributions for all cuts, not yet tested and very experimental\n",
        "distribution = 'none'#@param['none','random','fixed']\n",
        "\n",
        "if distribution == 'fixed':\n",
        "  use_bucketpow = True\n",
        "  use_rand_bucketpow = False \n",
        "elif distribution == 'random':\n",
        "  use_bucketpow = False\n",
        "  use_rand_bucketpow = True \n",
        "else:\n",
        "  use_bucketpow = False\n",
        "  use_rand_bucketpow = False \n",
        "\n",
        "\n",
        "\n",
        "bucket_cuts = 16 #@param{type:'integer'}\n",
        "\n",
        "smallcut_size = 0#@param{type:'number'}\n",
        "smallcut_p = 0.35 #@param{type:'number'}\n",
        "mediumcut_size = 0.08 #@param{type:'number'}\n",
        "mediumcut_p = 0.5 #@param{type:'number'}\n",
        "largecut_size = 0.2 #@param{type:'number'}\n",
        "largecut_p = 0.7 #@param{type:'number'}\n",
        "\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "cellView": "form",
        "id": "z1C6MMhhQQ44"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "### prompt randomizers"
      ],
      "metadata": {
        "id": "qr7anWTE0H2_"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import random\n",
        "\n",
        "randomizers = [\n",
        "\t[\"artist\", \"Asher Brown Durand\"],\n",
        "\t[\"artist\", \"Greg Rutkowski\"],\n",
        "\t[\"artist\", \"Michael Whelan\"],\n",
        "\t[\"artist\", \"Chriss Foss\"],\n",
        "\t[\"artist\", \"Marc Simonetti\"],\n",
        "\t[\"artist\", \"Anato Finnstark\"],\n",
        "\t[\"artist\", \"Jean Delville\"],\n",
        "\t[\"artist\", \"John Harris\"],\n",
        "\t[\"artist\", \"Simon Stalenhag\"],\n",
        "\t[\"artist\", \"John Howe\"],\n",
        "\t[\"artist\", \"James Jean\"],\n",
        "\t[\"artist\", \"Don Maitz\"],\n",
        "\t[\"artist\", \"Gediminas Pranckevicius\"],\n",
        "\t[\"artist\", \"Zdzisław Beksiński\"],\n",
        "\t[\"artist\", \"Albert Bierstadt\"],\n",
        "\t[\"artist\", \"Peter Mohrbacher\"],\n",
        "\t[\"artist\", \"John Atkinson Grimshaw\"],\n",
        "\t[\"artist\", \"Peter Gric\"],\n",
        "\t[\"artist\", \"Paul Delvaux\"],\n",
        "\t[\"artist\", \"Rob Gonsalves\"],\n",
        "\t[\"artist\", \"Karol Bak\"],\n",
        "\t[\"artist\", \"Hubert Robert\"],\n",
        "\t[\"artist\", \"Michael Parkes\"],\n",
        "\t[\"artist\", \"Andrew Wyeth\"],\n",
        "\t[\"artist\", \"Rebecca Guay\"],\n",
        "\t[\"artist\", \"Fairfield Porter\"],\n",
        "\t[\"artist\", \"Jacek Yerka\"],\n",
        "\t[\"artist\", \"John Berkey\"],\n",
        "\t[\"artist\", \"Margaret Macdonald Mackintosh\"],\n",
        "\t[\"artist\", \"Peter Elson\"],\n",
        "\t[\"artist\", \"Chris Foss\"],\n",
        "\t[\"artist\", \"Bob Eggleton\"],\n",
        "\t[\"artist\", \"Takato Yamamoto\"],\n",
        "\t[\"artist\", \"Odd Nerdrum\"],\n",
        "\t[\"artist\", \"Dean Cornwell\"],\n",
        "\t[\"artist\", \"Robert McCall\"],\n",
        "\t[\"artist\", \"Edmund Dulac\"],\n",
        "\t[\"artist\", \"George Inness\"],\n",
        "\t[\"artist\", \"Rodney Matthews\"],\n",
        "\t[\"artist\", \"Martin Johnson Heade\"],\n",
        "\t[\"artist\", \"Lee Madgwick\"],\n",
        "\t[\"artist\", \"Beksinski\"],\n",
        "\t[\"artist\", \"Gustave Moreau\"],\n",
        "\t[\"adjective\", \"beautiful\"],\n",
        "\t[\"adjective\", \"fascinating\"],\n",
        "\t[\"adjective\", \"interesting\"],\n",
        "\t[\"adjective\", \"dire\"],\n",
        "\t[\"adjective\", \"terrifying\"],\n",
        "\t[\"adjective\", \"elusive\"],\n",
        "\t[\"adjective\", \"frantic\"],\n",
        "\t[\"adjective\", \"serene\"],\n",
        "\t[\"adjective\", \"evil\"],\n",
        "\t[\"adjective\", \"festive\"],\n",
        "\t[\"adjective\", \"mind-bending\"],\n",
        "\t[\"adjective\", \"complex\"],\n",
        "\t[\"adjective\", \"simple\"],\n",
        "\t[\"adjective\", \"extraordinary\"],\n",
        "\t[\"adjective\", \"masterpiece\"],\n",
        "\t[\"adjective\", \"tangible\"],\n",
        "\t[\"adjective\", \"psychotic\"],\n",
        "\t[\"adjective\", \"warlike\"],\n",
        "\t[\"adjective\", \"unusual\"],\n",
        "\t[\"site\", \"trending on ArtStation\"],\n",
        "\t[\"site\", \"trending on ConceptArtWorld\"],\n",
        "\t[\"site\", \"trending on Flickr\"],\n",
        "\t[\"site\", \"trending on Behance\"],\n",
        "\t[\"site\", \"trending on 500px\"],\n",
        "\t[\"site\", \"trending on Getty Images\"],\n",
        "\t[\"site\", \"trending on DeviantArt\"],\n",
        "\t[\"site\", \"trending on Pixiv\"],\n",
        "\t[\"site\", \"trending on IllustrationX\"],\n",
        "\t[\"site\", \"trending on cgsociety\"]\n",
        "]\n",
        "\n",
        "def randomize(category):\n",
        "    random.seed()\n",
        "    choices = []\n",
        "    for item in randomizers:\n",
        "        if item[0] == category:\n",
        "            choices.append(item[1])\n",
        "    random_choice = random.choice(choices)\n",
        "    return (random_choice)\n",
        "\n",
        "\n",
        "def randomize_prompts(prompt):\n",
        "\tnewprompt = prompt\n",
        "\tif \"_\" in newprompt:\n",
        "\t\twhile \"_\" in newprompt:\n",
        "\t\t\tstart = newprompt.index('_')\n",
        "\t\t\tend = newprompt.index('_',start+1)\n",
        "\t\t\tswap = newprompt[(start + 1):end]\n",
        "\t\t\tswapped = randomize(swap)\n",
        "\t\t\tnewprompt = newprompt.replace(f'_{swap}_', swapped, 1)\n",
        "\treturn newprompt\n",
        "\n",
        "prompt_to_randomize = \"A _adjective_ painting by _artist_ trending on _site_\"#@param {type:'string'}\n",
        "iterations = 5 #@param {type:'integer'}\n",
        "\n",
        "for i in range (iterations):\n",
        "    print('[\"' + randomize_prompts(prompt_to_randomize) + '\"]')\n"
      ],
      "metadata": {
        "cellView": "form",
        "id": "P7U_t4so0JqQ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PromptsTop"
      },
      "source": [
        "### Prompts\n",
        "`animation_mode: None` will only use the first set. `animation_mode: 2D / Video` will run through them per the set frames and hold on the last one."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Prompts"
      },
      "source": [
        "# Note: If using a pixelart diffusion model, try adding \"#pixelart\" to the end of the prompt for a stronger effect. It'll tend to work a lot better!\n",
        "\n",
        "text_prompt_list = [\n",
        "    \n",
        "\n",
        "  [\"A beautiful painting of a singular lighthouse, shining its light across a tumultuous sea of blood by greg rutkowski and thomas kinkade, Trending on artstation:[5]*500+[1]*500\", \"yellow color scheme:[1]*500+[1]*500\"],\n",
        "  #[\"prompt i want to go after the first\"]\n",
        "\n",
        "# can now run a list of prompts, each prompt goes in a set of brackets, on a new line, and with a comma at the end of the brackets.\n",
        "\n",
        "]\n",
        "\n",
        "\n",
        "image_prompts = {\n",
        "    # 0:['ImagePromptsWorkButArentVeryGood.png:2',],\n",
        "}"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "code",
      "source": [
        "text_prompts_animation = {\n",
        "    \n",
        "    0: [\"big cool lighthouse\"],\n",
        "    100: [\"this prompt will start at the 100th frame of the animation, denoted by the number on the left\"],\n",
        "\n",
        "\n",
        "\n",
        "}"
      ],
      "metadata": {
        "id": "TVvEmoGlnBml"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Init Image Settings\n",
        "skip steps are now set with the init image, in this format [imgpath, skipsteps] see below"
      ],
      "metadata": {
        "id": "q1CVztDi48o3"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#similar to prompt list, but no need for brackets here. if number of init images is less than number of prompts, only the first will be used, otherwise, each will go together\n",
        "#i.e prompt1 with init_image1 followed by prompt2 with init_image2 and so on\n",
        "\n",
        "#the second number in the brackets indicates how many skip steps\n",
        "\n",
        "init_image_list = [\n",
        "    \n",
        " # [\"/content/pic1.jpeg\",50],\n",
        "\n",
        "\n",
        "\n",
        "]"
      ],
      "metadata": {
        "id": "N2hpMCAz5Ezw"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@markdown ####**Init Image Settings:**\n",
        "init_scale = 0 #@param{type: 'integer'}\n",
        "skip_steps_no_init = 0 #@param{type: 'integer'}\n",
        "#@markdown *Make sure you set skip_steps to ~50% of your steps if you want to use an init image.*\n"
      ],
      "metadata": {
        "cellView": "form",
        "id": "Gyrme5GO4_nG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DoTheRun",
        "cellView": "form"
      },
      "source": [
        "#@title #Do the Run!\n",
        "#@markdown `n_batches` ignored with animation modes.\n",
        "display_rate = 1 #param{type: 'number'}\n",
        "n_batches = 1 #@param{type: 'number'}\n",
        "\n",
        "\n",
        "# convert schedules in case of wrong format\n",
        "def convert_to_schedule(string):\n",
        "  if \"[\" not in string:\n",
        "    newstring = \"[\" + string + \"]*1000\"\n",
        "  else:\n",
        "    newstring = string\n",
        "\n",
        "  return newstring\n",
        "\n",
        "\n",
        "\n",
        "use_secondary_model = convert_to_schedule(use_secondary_model)\n",
        "\n",
        "clip_guidance_scale = convert_to_schedule(clip_guidance_scale)\n",
        "\n",
        "cutn_batches = convert_to_schedule(cutn_batches)\n",
        "\n",
        "clamp_max = convert_to_schedule(clamp_max)\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "if animation_mode == 'Video Input':\n",
        "    steps = video_init_steps\n",
        "\n",
        "#Update Model Settings\n",
        "timestep_respacing = f'ddim{steps}'\n",
        "diffusion_steps = (1000//steps)*steps if steps < 1000 else steps\n",
        "model_config.update({\n",
        "    'timestep_respacing': timestep_respacing,\n",
        "    'diffusion_steps': diffusion_steps,\n",
        "})\n",
        "\n",
        "batch_size = 1 \n",
        "\n",
        "def move_files(start_num, end_num, old_folder, new_folder):\n",
        "    for i in range(start_num, end_num):\n",
        "        old_file = old_folder + f'/{batch_name}({batchNum})_{i:04}.png'\n",
        "        new_file = new_folder + f'/{batch_name}({batchNum})_{i:04}.png'\n",
        "        os.rename(old_file, new_file)\n",
        "\n",
        "#@markdown ---\n",
        "\n",
        "\n",
        "resume_run = False #@param{type: 'boolean'}\n",
        "run_to_resume = 'latest' #@param{type: 'string'}\n",
        "resume_from_frame = 'latest' #@param{type: 'string'}\n",
        "retain_overwritten_frames = False #@param{type: 'boolean'}\n",
        "if retain_overwritten_frames:\n",
        "    retainFolder = f'{batchFolder}/retained'\n",
        "    createPath(retainFolder)\n",
        "\n",
        "\n",
        "skip_step_ratio = int(frames_skip_steps.rstrip(\"%\")) / 100\n",
        "calc_frames_skip_steps = math.floor(steps * skip_step_ratio)\n",
        "\n",
        "if animation_mode == 'Video Input':\n",
        "    frames = sorted(glob(in_path+'/*.*'));\n",
        "    if len(frames)==0: \n",
        "        sys.exit(\"ERROR: 0 frames found.\\nPlease check your video input path and rerun the video settings cell.\")\n",
        "    flows = glob(flo_folder+'/*.*')\n",
        "    if (len(flows)==0) and video_init_flow_warp:\n",
        "        sys.exit(\"ERROR: 0 flow files found.\\nPlease rerun the flow generation cell.\")\n",
        "\n",
        "if steps <= calc_frames_skip_steps:\n",
        "    sys.exit(\"ERROR: You can't skip more steps than your total steps\")\n",
        "\n",
        "if resume_run:\n",
        "    if run_to_resume == 'latest':\n",
        "        try:\n",
        "            batchNum\n",
        "        except:\n",
        "            batchNum = len(glob(f\"{batchFolder}/{batch_name}(*)_settings.txt\"))-1\n",
        "    else:\n",
        "        batchNum = int(run_to_resume)\n",
        "    if resume_from_frame == 'latest':\n",
        "        start_frame = len(glob(batchFolder+f\"/{batch_name}({batchNum})_*.png\"))\n",
        "        if animation_mode != '3D' and turbo_mode == True and start_frame > turbo_preroll and start_frame % int(turbo_steps) != 0:\n",
        "            start_frame = start_frame - (start_frame % int(turbo_steps))\n",
        "    else:\n",
        "        start_frame = int(resume_from_frame)+1\n",
        "        if animation_mode != '3D' and turbo_mode == True and start_frame > turbo_preroll and start_frame % int(turbo_steps) != 0:\n",
        "            start_frame = start_frame - (start_frame % int(turbo_steps))\n",
        "        if retain_overwritten_frames is True:\n",
        "            existing_frames = len(glob(batchFolder+f\"/{batch_name}({batchNum})_*.png\"))\n",
        "            frames_to_save = existing_frames - start_frame\n",
        "            print(f'Moving {frames_to_save} frames to the Retained folder')\n",
        "            move_files(start_frame, existing_frames, batchFolder, retainFolder)\n",
        "else:\n",
        "    start_frame = 0\n",
        "    batchNum = len(glob(batchFolder+\"/*.txt\"))\n",
        "    while os.path.isfile(f\"{batchFolder}/{batch_name}({batchNum})_settings.txt\") or os.path.isfile(f\"{batchFolder}/{batch_name}-{batchNum}_settings.txt\"):\n",
        "        batchNum += 1\n",
        "\n",
        "\n",
        "\n",
        "if animation_mode != 'None':\n",
        "    text_prompt_list = ['hi']\n",
        "\n",
        "for i, prompt in enumerate(text_prompt_list):\n",
        "  print(f'Starting Run: {batch_name}({batchNum}) at frame {start_frame}')\n",
        "\n",
        "  if animation_mode == 'None':\n",
        "    text_prompts = {0:prompt}\n",
        "    if (len(init_image_list) >= len(text_prompt_list)):\n",
        "      init_image = init_image_list[i][0]\n",
        "      skip_steps = init_image_list[i][1]\n",
        "    else:\n",
        "      try:\n",
        "        init_image = init_image_list[0][0]\n",
        "        skip_steps = init_image_list[0][1]\n",
        "      except:\n",
        "        init_image = None\n",
        "        skip_steps = skip_steps_no_init\n",
        "    \n",
        "    if init_image:\n",
        "      perlin_init = False\n",
        "\n",
        "  else:\n",
        "    text_prompts = text_prompts_animation\n",
        "    try:\n",
        "      init_image = init_image_list[0]\n",
        "    except:\n",
        "      init_image = None\n",
        "    if init_image:\n",
        "      perlin_init = False\n",
        "  \n",
        "  if animation_mode == 'Video Input':\n",
        "      skip_steps = video_init_skip_steps\n",
        "\n",
        "\n",
        "  if set_seed == 'random_seed' or '':\n",
        "      random.seed()\n",
        "      seed = random.randint(0, 2**32)\n",
        "      # print(f'Using seed: {seed}')\n",
        "  else:\n",
        "      seed = int(set_seed)\n",
        "\n",
        "\n",
        "  args = {\n",
        "      'batchNum': batchNum,\n",
        "      'prompts_series':split_prompts(text_prompts) if text_prompts else None,\n",
        "      'image_prompts_series':split_prompts(image_prompts) if image_prompts else None,\n",
        "      'seed': seed,\n",
        "      'display_rate':display_rate,\n",
        "      'n_batches':n_batches if animation_mode == 'None' else 1,\n",
        "      'batch_size':batch_size,\n",
        "      'batch_name': batch_name,\n",
        "      'steps': steps,\n",
        "      'diffusion_sampling_mode': diffusion_sampling_mode,\n",
        "      'width_height': width_height,\n",
        "      'clip_guidance_scale': eval(clip_guidance_scale),\n",
        "      'tv_scale': tv_scale,\n",
        "      'range_scale': range_scale,\n",
        "      'sat_scale': sat_scale,\n",
        "      'cutn_batches': eval(cutn_batches),\n",
        "      'init_image': init_image,\n",
        "      'init_scale': init_scale,\n",
        "      'skip_steps': skip_steps,\n",
        "      'side_x': side_x,\n",
        "      'side_y': side_y,\n",
        "      'timestep_respacing': timestep_respacing,\n",
        "      'diffusion_steps': diffusion_steps,\n",
        "      'animation_mode': animation_mode,\n",
        "      'video_init_path': video_init_path,\n",
        "      'extract_nth_frame': extract_nth_frame,\n",
        "      'video_init_seed_continuity': video_init_seed_continuity,\n",
        "      'key_frames': key_frames,\n",
        "      'max_frames': max_frames if animation_mode != \"None\" else 1,\n",
        "      'interp_spline': interp_spline,\n",
        "      'start_frame': start_frame,\n",
        "      'angle': angle,\n",
        "      'zoom': zoom,\n",
        "      'translation_x': translation_x,\n",
        "      'translation_y': translation_y,\n",
        "      'translation_z': translation_z,\n",
        "      'rotation_3d_x': rotation_3d_x,\n",
        "      'rotation_3d_y': rotation_3d_y,\n",
        "      'rotation_3d_z': rotation_3d_z,\n",
        "      'midas_depth_model': midas_depth_model,\n",
        "      'midas_weight': midas_weight,\n",
        "      'near_plane': near_plane,\n",
        "      'far_plane': far_plane,\n",
        "      'fov': fov,\n",
        "      'padding_mode': padding_mode,\n",
        "      'sampling_mode': sampling_mode,\n",
        "      'angle_series':angle_series,\n",
        "      'zoom_series':zoom_series,\n",
        "      'translation_x_series':translation_x_series,\n",
        "      'translation_y_series':translation_y_series,\n",
        "      'translation_z_series':translation_z_series,\n",
        "      'rotation_3d_x_series':rotation_3d_x_series,\n",
        "      'rotation_3d_y_series':rotation_3d_y_series,\n",
        "      'rotation_3d_z_series':rotation_3d_z_series,\n",
        "      'frames_scale': frames_scale,\n",
        "      'skip_step_ratio': skip_step_ratio,\n",
        "      'calc_frames_skip_steps': calc_frames_skip_steps,\n",
        "      'text_prompts': text_prompts,\n",
        "      'image_prompts': image_prompts,\n",
        "      'use_secondary_model': eval(use_secondary_model),\n",
        "      'cut_overview': eval(cut_overview),\n",
        "      'cut_innercut': eval(cut_innercut),\n",
        "      'cut_ic_pow': eval(cut_ic_pow),\n",
        "      'ic_gray_p': eval(ic_gray_p),\n",
        "      'ov_gray_p': eval(ov_gray_p),\n",
        "      'percent_overview': eval(percent_overview),\n",
        "      'floor_size': eval(floor_size),\n",
        "      'overview_weight':eval(overview_weight),\n",
        "      'cut_weight_multiplier': eval(cut_weight_multiplier),\n",
        "      'bucket_cuts': bucket_cuts,\n",
        "      'ceiling': ceiling,\n",
        "      'intermediate_saves': intermediate_saves,\n",
        "      'intermediates_in_subfolder': intermediates_in_subfolder,\n",
        "      'steps_per_checkpoint': steps_per_checkpoint,\n",
        "      'perlin_init': perlin_init,\n",
        "      'perlin_mode': perlin_mode,\n",
        "      'set_seed': set_seed,\n",
        "      'eta': eta,\n",
        "      'clamp_grad': clamp_grad,\n",
        "      'clamp_max': eval(clamp_max),\n",
        "      'skip_augs': skip_augs,\n",
        "      'randomize_class': randomize_class,\n",
        "      'clip_denoised': clip_denoised,\n",
        "      'fuzzy_prompt': fuzzy_prompt,\n",
        "      'rand_mag': rand_mag,\n",
        "      'turbo_mode':turbo_mode,\n",
        "      'turbo_steps':turbo_steps,\n",
        "      'turbo_preroll':turbo_preroll,\n",
        "      'resize_mid_run': resize_mid_run,\n",
        "      'use_vertical_symmetry': use_vertical_symmetry,\n",
        "      'use_horizontal_symmetry': use_horizontal_symmetry,\n",
        "      'transformation_percent': transformation_percent,\n",
        "      #video init settings\n",
        "      'video_init_steps': video_init_steps,\n",
        "      'video_init_clip_guidance_scale': video_init_clip_guidance_scale,\n",
        "      'video_init_tv_scale': video_init_tv_scale,\n",
        "      'video_init_range_scale': video_init_range_scale,\n",
        "      'video_init_sat_scale': video_init_sat_scale,\n",
        "      'video_init_cutn_batches': video_init_cutn_batches,\n",
        "      'video_init_skip_steps': video_init_skip_steps,\n",
        "      'video_init_frames_scale': video_init_frames_scale,\n",
        "      'video_init_frames_skip_steps': video_init_frames_skip_steps,\n",
        "      #warp settings\n",
        "      'video_init_flow_warp':video_init_flow_warp,\n",
        "      'video_init_flow_blend':video_init_flow_blend,\n",
        "      'video_init_check_consistency':video_init_check_consistency,\n",
        "      'video_init_blend_mode':video_init_blend_mode,\n",
        "      'symmetry_loss':symmetry_loss,\n",
        "      'sloss_scale':symm_loss_scale,\n",
        "      'symm_switch':symm_switch,\n",
        "      'noise_injector': noise_injector,\n",
        "      'noise_injector_threshold': noise_injector_threshold,\n",
        "      'noise_injector_mix': noise_injector_mix,\n",
        "      'noise_injector_px_size': noise_injector_px_size,\n",
        "      'noise_injector_spice': noise_injector_spice,\n",
        "      'noise_injector_mask_feather': noise_injector_mask_feather,\n",
        "      'noise_injector_sigma': noise_injector_sigma,\n",
        "      'copy_palette': copy_palette,\n",
        "      'copy_palette_image': copy_palette_image,\n",
        "  }\n",
        "\n",
        "\n",
        "\n",
        "  if animation_mode == 'Video Input':\n",
        "      # This isn't great in terms of what will get saved to the settings.. but it should work.\n",
        "      args['steps'] = args['video_init_steps']\n",
        "      args['clip_guidance_scale'] = args['video_init_clip_guidance_scale']\n",
        "      args['tv_scale'] = args['video_init_tv_scale']\n",
        "      args['range_scale'] = args['video_init_range_scale']\n",
        "      args['sat_scale'] = args['video_init_sat_scale']\n",
        "      args['cutn_batches'] = args['video_init_cutn_batches']\n",
        "      args['skip_steps'] = args['video_init_skip_steps']\n",
        "      args['frames_scale'] = args['video_init_frames_scale']\n",
        "      args['frames_skip_steps'] = args['video_init_frames_skip_steps']\n",
        "\n",
        "  args = SimpleNamespace(**args)\n",
        "\n",
        "  print('Prepping model...')\n",
        "  model, diffusion = create_model_and_diffusion(**model_config)\n",
        "  if diffusion_model == 'custom':\n",
        "      model.load_state_dict(torch.load(custom_path, map_location='cpu'))\n",
        "  else:\n",
        "      model.load_state_dict(torch.load(f'{model_path}/{get_model_filename(diffusion_model)}', map_location='cpu'))\n",
        "  model.requires_grad_(False).eval().to(device)\n",
        "  for name, param in model.named_parameters():\n",
        "      if 'qkv' in name or 'norm' in name or 'proj' in name:\n",
        "          param.requires_grad_()\n",
        "  if model_config['use_fp16']:\n",
        "      model.convert_to_fp16()\n",
        "\n",
        "  gc.collect()\n",
        "  torch.cuda.empty_cache()\n",
        "\n",
        "\n",
        "  try:\n",
        "      do_run()\n",
        "      batchNum += 1\n",
        "  except KeyboardInterrupt:\n",
        "      pass\n",
        "  finally:\n",
        "      print('Seed used:', seed)\n",
        "      gc.collect()\n",
        "      torch.cuda.empty_cache()\n"
      ],
      "outputs": [],
      "execution_count": null
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CreateVidTop"
      },
      "source": [
        "# 5. Create the video"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CreateVid",
        "cellView": "form"
      },
      "source": [
        "import PIL\n",
        "# @title ### **Create video**\n",
        "#@markdown Video file will save in the same folder as your images.\n",
        "from tqdm.notebook import trange\n",
        "skip_video_for_run_all = True #@param {type: 'boolean'}\n",
        "\n",
        "if animation_mode == 'Video Input':\n",
        "    frames = sorted(glob(in_path+'/*.*'));\n",
        "    if len(frames)==0: \n",
        "        sys.exit(\"ERROR: 0 frames found.\\nPlease check your video input path and rerun the video settings cell.\")\n",
        "    flows = glob(flo_folder+'/*.*')\n",
        "    if (len(flows)==0) and video_init_flow_warp:\n",
        "        sys.exit(\"ERROR: 0 flow files found.\\nPlease rerun the flow generation cell.\")\n",
        "\n",
        "blend =  0.5#@param {type: 'number'}\n",
        "video_init_check_consistency = False #@param {type: 'boolean'}\n",
        "if skip_video_for_run_all == True:\n",
        "    print('Skipping video creation, uncheck skip_video_for_run_all if you want to run it')\n",
        "\n",
        "else:\n",
        "    # import subprocess in case this cell is run without the above cells\n",
        "    import subprocess\n",
        "    from base64 import b64encode\n",
        "\n",
        "    latest_run = batchNum\n",
        "\n",
        "    folder = \"batch_name\" #@param {type: 'string'}\n",
        "    run = latest_run #@param\n",
        "    final_frame = 'final_frame'\n",
        "\n",
        "\n",
        "    init_frame = 1#@param {type:\"number\"} This is the frame where the video will start\n",
        "    last_frame = final_frame#@param {type:\"number\"} You can change i to the number of the last frame you want to generate. It will raise an error if that number of frames does not exist.\n",
        "    fps = 12#@param {type:\"number\"}\n",
        "    # view_video_in_cell = True #@param {type: 'boolean'}\n",
        "\n",
        "    frames = []\n",
        "    # tqdm.write('Generating video...')\n",
        "\n",
        "    if last_frame == 'final_frame':\n",
        "        last_frame = len(glob(batchFolder+f\"/{folder}({run})_*.png\"))\n",
        "        print(f'Total frames: {last_frame}')\n",
        "\n",
        "    image_path = f\"{outDirPath}/{folder}/{folder}({run})_%04d.png\"\n",
        "    filepath = f\"{outDirPath}/{folder}/{folder}({run}).mp4\"\n",
        "\n",
        "    if (video_init_blend_mode == 'optical flow') and (animation_mode == 'Video Input'):\n",
        "        image_path = f\"{outDirPath}/{folder}/flow/{folder}({run})_%04d.png\"\n",
        "        filepath = f\"{outDirPath}/{folder}/{folder}({run})_flow.mp4\"\n",
        "        if last_frame == 'final_frame':\n",
        "            last_frame = len(glob(batchFolder+f\"/flow/{folder}({run})_*.png\"))\n",
        "        flo_out = batchFolder+f\"/flow\"\n",
        "        createPath(flo_out)\n",
        "        frames_in = sorted(glob(batchFolder+f\"/{folder}({run})_*.png\"))\n",
        "        shutil.copy(frames_in[0], flo_out)\n",
        "        for i in trange(init_frame, min(len(frames_in), last_frame)):\n",
        "            frame1_path = frames_in[i-1]\n",
        "            frame2_path = frames_in[i]\n",
        "  \n",
        "            frame1 = PIL.Image.open(frame1_path)\n",
        "            frame2 = PIL.Image.open(frame2_path)\n",
        "            frame1_stem = f\"{(int(frame1_path.split('/')[-1].split('_')[-1][:-4])+1):04}.jpg\"\n",
        "            flo_path = f\"/{flo_folder}/{frame1_stem}.npy\"\n",
        "            weights_path = None\n",
        "            if video_init_check_consistency:\n",
        "                # TBD\n",
        "                pass\n",
        "            warp(frame1, frame2, flo_path, blend=blend, weights_path=weights_path).save(batchFolder+f\"/flow/{folder}({run})_{i:04}.png\")\n",
        "    if video_init_blend_mode == 'linear':\n",
        "        image_path = f\"{outDirPath}/{folder}/blend/{folder}({run})_%04d.png\"\n",
        "        filepath = f\"{outDirPath}/{folder}/{folder}({run})_blend.mp4\"\n",
        "        if last_frame == 'final_frame':\n",
        "            last_frame = len(glob(batchFolder+f\"/blend/{folder}({run})_*.png\"))\n",
        "        blend_out = batchFolder+f\"/blend\"\n",
        "        createPath(blend_out)\n",
        "        frames_in = glob(batchFolder+f\"/{folder}({run})_*.png\")\n",
        "        shutil.copy(frames_in[0], blend_out)\n",
        "        for i in trange(1, len(frames_in)):\n",
        "            frame1_path = frames_in[i-1]\n",
        "            frame2_path = frames_in[i]\n",
        "    \n",
        "            frame1 = PIL.Image.open(frame1_path)\n",
        "            frame2 = PIL.Image.open(frame2_path)\n",
        "          \n",
        "            frame = PIL.Image.fromarray((np.array(frame1)*(1-blend) + np.array(frame2)*(blend)).astype('uint8')).save(batchFolder+f\"/blend/{folder}({run})_{i:04}.png\")\n",
        "\n",
        "\n",
        "    cmd = [\n",
        "        'ffmpeg',\n",
        "        '-y',\n",
        "        '-vcodec',\n",
        "        'png',\n",
        "        '-r',\n",
        "        str(fps),\n",
        "        '-start_number',\n",
        "        str(init_frame),\n",
        "        '-i',\n",
        "        image_path,\n",
        "        '-frames:v',\n",
        "        str(last_frame+1),\n",
        "        '-c:v',\n",
        "        'libx264',\n",
        "        '-vf',\n",
        "        f'fps={fps}',\n",
        "        '-pix_fmt',\n",
        "        'yuv420p',\n",
        "        '-crf',\n",
        "        '17',\n",
        "        '-preset',\n",
        "        'veryslow',\n",
        "        filepath\n",
        "    ]\n",
        "\n",
        "    process = subprocess.Popen(cmd, cwd=f'{batchFolder}', stdout=subprocess.PIPE, stderr=subprocess.PIPE)\n",
        "    stdout, stderr = process.communicate()\n",
        "    if process.returncode != 0:\n",
        "        print(stderr)\n",
        "        raise RuntimeError(stderr)\n",
        "    else:\n",
        "        print(\"The video is ready and saved to the images folder\")\n",
        "\n",
        "    # if view_video_in_cell:\n",
        "    #     mp4 = open(filepath,'rb').read()\n",
        "    #     data_url = \"data:video/mp4;base64,\" + b64encode(mp4).decode()\n",
        "    #     display.HTML(f'<video width=400 controls><source src=\"{data_url}\" type=\"video/mp4\"></video>')"
      ],
      "outputs": [],
      "execution_count": null
    }
  ],
  "metadata": {
    "anaconda-cloud": {},
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [
        "CreditsChTop",
        "TutorialTop",
        "NY8nN3u4SE5a",
        "AnimSetTop",
        "ExtraSetTop",
        "wTJTMhofQNQW",
        "DiffuseTop"
      ],
      "machine_shape": "hm",
      "private_outputs": true,
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.1"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}