Pixart Sigma - ComfyUI
Added 2024-04-20 14:20:01 +0000 UTC
Pixart Sigma - https://pixart-alpha.github.io/PixArt-sigma-project/
We're seeing more T5 conditioning over CLIP, often with much better prompt adherence. This time, it's the recent release "Pixart Sigma".
Command summary
Use your own workspace and python environment:
cd your-workspace
conda activate comfyui
cd ComfyUI
git clone https://github.com/city96/ComfyUI_ExtraModels custom_nodes/ComfyUI_ExtraModels
cd custom_nodes/ComfyUI_ExtraModels
pip install -r requirements.txt
cd your-workspace
git clone https://github.com/PixArt-alpha/PixArt-sigma.git
cd PixArt-sigma
pip install -r requirements.txt
pip install evaluate
git lfs install
git clone https://huggingface.co/PixArt-alpha/pixart_sigma_sdxlvae_T5_diffusers output/pretrained_models/pixart_sigma_sdxlvae_T5_diffusers
python tools/download.py
Once downloaded, move models to
ComfyUI/models/checkpoints/PixArt-Sigma-XL-2-1024-MS.pth
ComfyUI/models/t5/pixart_sigma_sdxlvae_T5_diffusers
ComfyUI/models/VAE/sd-vae-ft-ema
Update
Added a Pixart Sigma Refiner variant which uses SDXL to change the image slightly, which can often improve the visual aesthetic.
Update 2
Added an even more refining + upscaling option ;)
