Majesty Diffusion 👑

Generate images from text with majesty

Formerly known as Princess Generator

Majesty Diffusion are implementations of text-to-image diffusion models with a royal touch 👸

Access our Majestic Guide (under construction), join our community on Discord or reach out via @multimodalart on Twitter). Also share your settings with us

Current implementations:

• Latent Majesty Diffusion
• V-Majesty Diffusion

Latent Majesty Diffusion v1.6

Formerly known as Latent Princess Generator

A Dango233 and apolinario (@multimodalart) Colab notebook implementing CompVis' Latent Diffusion. Contribute to our settings library on Hugging Face!

v1.2

• Added Dango233 CLIP Guidance
• Added Dango233 magical new step and upscaling scheduling
• Added Dango233 cuts, augs and attributes scheduling
• Added Dango233 mag and clamp settings
• Added Dango233 linear ETA scheduling
• Added Dango233 negative prompts for Latent Diffusion Guidance
• Added Jack000 GLID-3 XL watermark free fine-tuned model
• Added dmarx Multi-Modal-Comparators for CLIP and CLIP-like models
• Added open_clip gradient checkpointing
• Added crowsonkb aesthetic models
• Added LAION-AI aesthetic predictor embeddings
• Added Dango233 inpainting mode
• Added apolinario (@multimodalart) savable settings and setting library (including colab-free-default, dango233-princesses, the-other-zippy and makaitrad shared settings. Share yours with us too with a pull request!

v1.3

- Better Upscaler (learn how to use it on our [Majestic Guide](https://multimodal.art/majesty-diffusion))

v1.4 & 1.5 & 1.6

v1.4

• Added Dango233 Customised Dynamic Thresholding
• Added open_clip ViT-L/14 LAION-400M trained
• Fix CLOOB perceptor from MMC
• Removes latent upscaler (was broken), adds RGB upscaler

v1.5

• Even better defaults
• Better dynamic thresholidng
• Improves range scale
• Adds var scale and mean scale
• Adds the possibility of blurring cuts
• Adds experimental compression and punishment settings
• Adds PLMS support (experimental, results perceptually weird)

v1.6

• Adds LAION ongo (finetuned in artworks) and erlich (finetuned for making logos) models
• Adds noising and scaling during the advanced schedulign phases
• Adds ViT-L conditioning downstream to the Latent Diffusion unet process
• Small tweaks on dynamic thresholding
• Fixes linear ETA

V-Majesty Diffusion v1.2

Formerly known as Princess Generator ver. Victoria

A Dango233 and apolinario (@multimodalart) Colab notebook implementing crowsonkb's V-Objective Diffusion, with the following changes:

• Added Dango233 parallel multi-model diffusion (e.g.: run cc12m_1 and yfcc_2 at the same time - with or without lerping)
• Added Dango233 cuts, augs and attributes scheduling
• Added Dango233 mag and clamp settings
• Added apolinario (@multimodalart) ETA scheduling
• Added nshepperd v-diffusion imagenet512 and danbooru models
• Added dmarx Multi-Modal-Comparators
• Added crowsonkb AVA and Simulacra bot aesthetic models
• Added LAION-AI aesthetic pre-calculated embeddings
• Added open_clip gradient checkpointing
• Added Dango233 inpainting mode
• Added apolinario (@multimodalart) "internal upscaling" (upscales the output with yfcc_2 or openimages)
• Added apolinario (@multimodalart) savable settings and setting library (including defaults, disco-diffusion-defaults default settings). Share yours with us too with a pull request!

TODO

Please feel free to help us in any of these tasks!

• Figure out better defaults and add more settings to the settings library (contribute with a PR!)
• Add all notebooks to a single pipeline where on model can be the output of the other (similar to Centipede Diffusion)
• Add all notebooks to the MindsEye UI
• Modularise everything
• Create a command line version
• Add an inpainting UI
• Improve performance, both in speed and VRAM consumption
• More technical issues will be listed on https://github.com/multimodalart/majesty-diffusion/issues

Acknowledgments

Some functions and methods are from various code masters - including but not limited to advadnoun, crowsonkb, nshepperd, russelldc, Dango233 and many others