This repository provides a ComfyUI custom node that wraps the Kwai-Kolors text-to-image diffusion model, enabling its use within the ComfyUI ecosystem. It is designed for users of ComfyUI who wish to leverage the capabilities of the Kwai-Kolors model for image generation.

How It Works

The wrapper utilizes the diffusers library to load and run the Kwai-Kolors text-to-image pipeline. It supports loading models from Hugging Face, including an alternative method for loading ChatGLM3 models from a single safetensors file. The base SDXL VAE is used for the process.

Quick Start & Requirements

• Install by cloning the repository into the ComfyUI/custom_nodes folder.
• Install dependencies: pip install -r requirements.txt (requires transformers version 4.38.0 or higher).
• Models (16.5GB fp16) are automatically downloaded to ComfyUI/models/diffusers/Kolors.
• The text encoder requires significant VRAM, with quantization options: fp16 (~13 GB), quant8 (~8 GB), quant4 (~4 GB).
• Image sampling at 1024 resolution is comparable to SDXL in VRAM usage.

Highlighted Details

• Supports loading quantized ChatGLM3 models from safetensors.
• Automatically downloads fp16 Kwai-Kolors models.
• Text encoder quantization options for reduced VRAM usage.
• Utilizes the base SDXL VAE.

Maintenance & Community

No specific information on maintainers, community channels, or roadmap is provided in the README.

Licensing & Compatibility

The README does not explicitly state the license for this wrapper. Compatibility with commercial or closed-source projects is not specified.

Limitations & Caveats

The text encoder is a significant VRAM consumer, even with quantization. The README does not detail specific performance benchmarks or known issues.