kijai / ComfyUI-KwaiKolorsWrapper

Diffusers wrapper to run Kwai-Kolors model
Apache License 2.0
551 stars 29 forks source link

ComfyUI wrapper for Kwai-Kolors

Rudimentary wrapper that runs Kwai-Kolors text2image pipeline using diffusers.

Update - safetensors

Added alternative way to load the ChatGLM3 model from single safetensors file (the configs are included in this repo already). Including already quantized models:

image

https://huggingface.co/Kijai/ChatGLM3-safetensors/upload/main

goes into:

ComfyUI\models\LLM\checkpoints image

image

Installation:

Clone this repository to 'ComfyUI/custom_nodes` folder.

Install the dependencies in requirements.txt, transformers version 4.38.0 minimum is required:

pip install -r requirements.txt

or if you use portable (run this in ComfyUI_windows_portable -folder):

python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-KwaiKolorsWrapper\requirements.txt

Models (fp16, 16.5GB) are automatically downloaded from https://huggingface.co/Kwai-Kolors/Kolors/tree/main

to ComfyUI/models/diffusers/Kolors

Model folder structure needs to be the following:

PS C:\ComfyUI_windows_portable\ComfyUI\models\diffusers\Kolors> tree /F
│   model_index.json
│
├───scheduler
│       scheduler_config.json
│
├───text_encoder
│       config.json
│       pytorch_model-00001-of-00007.bin
│       pytorch_model-00002-of-00007.bin
│       pytorch_model-00003-of-00007.bin
│       pytorch_model-00004-of-00007.bin
│       pytorch_model-00005-of-00007.bin
│       pytorch_model-00006-of-00007.bin
│       pytorch_model-00007-of-00007.bin
│       pytorch_model.bin.index.json
│       tokenizer.model
│       tokenizer_config.json
│       vocab.txt
│
└───unet
        config.json
        diffusion_pytorch_model.fp16.safetensors

To run this, the text enconder is what takes most of the VRAM, but can be quantized to fit approximately these amounts:

Model Size
fp16 ~13 GB
quant8 ~8 GB
quant4 ~4 GB

After that, the sampling single image at 1024 can be expected to take similar amounts than SDXL. For VAE the base SDXL VAE is used.

image

image