Kolors is an advanced text-to-image generation model leveraging latent diffusion, developed by the Kuaishou Kolors team. Trained on billions of text-image pairs, Kolors excels in visual quality, complex semantic understanding, and text rendering, outperforming both open-source and closed-source models. This model supports both Chinese and English inputs, making it highly effective in generating content specific to these languages.
Kolors has undergone extensive evaluation, using a dataset named KolorsPrompts, which includes over 1,000 prompts across 14 categories and 12 evaluation dimensions. The evaluation process involves both human and machine assessments. In human assessments, Kolors achieved the highest overall satisfaction score, significantly leading in visual appeal compared to other models. Machine assessments using the Multi-dimensional Human Preference Score (MPS) also indicated that Kolors achieved the highest scores, aligning with human evaluation results.
The model supports various functionalities, including IP-Adapter, ControlNet, Inpainting, and Dreambooth-LoRA, with detailed instructions for usage and inference provided. Kolors is open-sourced under the Apache-2.0 license, promoting collaboration within the open-source community while ensuring adherence to licensing terms for commercial use.
Pricing
Pricing information is not available