Kolors

Kolors generates photorealistic images from text for creators and developers who want to run image generation on their own hardware. Developed by Kuaishou, it supports Chinese and English prompts, including descriptions of Chinese cultural content.

• Bilingual generation: Describe scenes in Chinese or English and generate images containing text in either language.
• Reference images: Use IP-Adapter-Plus to guide generation with an image, or IP-Adapter-FaceID-Plus for face-based guidance.
• Composition control: Guide image structure with ControlNet using Canny edges, depth, or pose information.
• Editing and customization: Inpainting changes selected image regions, while Dreambooth-LoRA supports training and using custom adaptations of the model.

Downloadable model weights and inference code support local use. Kolors works with ComfyUI, Gradio, and Hugging Face Diffusers, giving users a choice between visual workflows and Python applications. Its Python and PyTorch stack recommends CUDA for GPU acceleration. The repository code is licensed under Apache 2.0.

The Kolors checkpoints are available through the official model downloads. The code uses Apache 2.0, while weights use the separate Kolors Model License. The project permits academic research and requires commercial users to register with the licensor and satisfy the model-license conditions.