This is an implementation of Ultimate-SD-Upscale via Comfy UI as a Cog model. Cog packages machine learning models as standard containers.
Model can be run here: https://replicate.com/fewjative/ultimate-sd-upscale
Initial logic to use ComfyUI is thanks to @Lucataco https://github.com/lucataco/cog-comfyui-sdxl-txt2img
First, download the models. There are 4 upscalers, 1 base model, and 1 controlnet model:
These go into 'upscaler-cache'
https://openmodeldb.info/models/4x-realesrgan-x4plus
https://openmodeldb.info/models/4x-realesrgan-x4plus-anime-6b
https://openmodeldb.info/models/4x-NMKD-Siax-CX
https://civitai.com/models/116225/4x-ultrasharp
These go into 'model-cache'
https://huggingface.co/XpucT/Deliberate/blob/main/Deliberate_v2.safetensors
These go into 'controlnet-cache'
https://huggingface.co/lllyasviel/ControlNet-v1-1/tree/main
This also requires ComfyUI. git clone this repo in the src dir: https://github.com/comfyanonymous/ComfyUI
And follow the instructions for the UltimateSDUpscale extension ( git clone in the ComfyUI/custom_nodes folders): https://github.com/ssitu/ComfyUI_UltimateSDUpscale
Lastly,in order to use the cache folder, you must modify this file to add new search entry points. Note you won't see this file until you clone ComfyUI: \cog-ultimate-sd-upscale\ComfyUI\extra_model_paths.yaml
other_ui: base_path: /src checkpoints: model-cache/ upscale_models: upscaler-cache/ controlnet: controlnet-cache/
Then, you can run predictions like such:
cog predict -i image=@toupscale.png
cog predict -i image=@jesko.png -i positive_prompt="A car from need for speed, in a garage, cinematic"
The workflow for ControlNet Tile and non Controlnet Tile Ultimate SD Upscale used for this repo is found under:
custom_workflows/ultimatesdupscale.json
custom_workflows/ultimatesdupscalecontrolnet.json
vs
