Learn about this first
OpenMOSS, together with Fudan NLP, Shanghai Chuangzhi, and MOSI, has released the open-weight image generation model MOSS-Image. Both the Base and SFT weights are licensed under Apache-2.0. The model was pretrained from scratch as a 5B-parameter single-stream DiT and is designed for tasks such as posters, infographics, comics, and UI sketches that require handling large amounts of text and spatial relationships among multiple elements.
According to the developers, MOSS-Image supports prompts of up to approximately 4K tokens, long-text rendering in Chinese and English, and directly editable JSON Layout. It can be used with Diffusers, Gradio, and ComfyUI, and supports LoRA and full fine-tuning. These capabilities are described by the developers and have not yet been verified by independent evaluations.
AI-generated from reports · updated 19 days agoLatest developments
OpenMOSS releases open-source image generation model MOSS-Image, focused on long-text rendering and editable layouts