AI Image Generator & Editor

Kolors AI is a latent diffusion model designed for photorealistic text-to-image synthesis, developed by the Kolors Team at Kuaishou Technology.

Kolors AI uses the General Language Model (GLM) to effectively comprehend and render both English and Chinese texts, ensuring accurate and detailed image generation.

Kolors AI's training strategy is divided into two phases: concept learning with broad knowledge and quality improvement with high-aesthetic data, optimizing the model's performance in generating high-quality images.

The code and weights for Kolors AI are available on GitHub at https://github.com/Kwai-Kolors/Kolors.

Yes, Kolors AI is free to use here. You can check the playground of this site.

Kolors AI is trained using a combination of public datasets like LAION, DataComp, and JourneyDB, as well as proprietary datasets. These datasets provide extensive coverage of visual concepts and high-aesthetic images.

Kolors AI employs a novel noise schedule and a two-phase training strategy to optimize high-resolution image generation. This approach enhances the visual appeal and detail of the generated images.

KolorsPrompts is a category-balanced benchmark created to guide the training and evaluation of Kolors AI. It includes prompts from various categories and challenges, ensuring comprehensive performance assessment.