MOSS-Image

Generate images with precise control over text and layout

Visit website

About MOSS-Image

MOSS-Image is an open-source family of image generation models that turns written descriptions into illustrations, photographs, and visual compositions. Its Base and SFT versions share a five-billion-parameter backbone and support prompts of up to 4,000 tokens, including instructions for rendering text inside images.

For compositions that need a specific arrangement, users can supply a JSON layout describing objects, words, and bounding boxes. This makes it possible to plan a poster or other graphic before generating the image. Developers can run the models through a Python Diffusers pipeline, a local Gradio interface, or ComfyUI, and adapt them with fine-tuning or LoRA training. Code and model weights are publicly available under Apache 2.0, with separate licenses applying to third-party components.

Discussion

Tried it? Share a useful thought or ask a question.