OpenMOSS Team releases Moss-Image AI model

The OpenMOSS Team has released Moss-Image, a new 5B parameter open-weight image generation AI model. Developed by research institutions in Shanghai with administrative backing from the Shanghai government, the model is designed to handle various artistic styles, ranging from photorealistic to anime.

Notably, the model demonstrates high linguistic capability, including the ability to render Japanese text accurately within generated images. It was trained on a massive dataset of 3.23 billion real images, specifically excluding AI-generated content to improve accuracy.

Moss-Image is released under the Apache 2.0 license, allowing for unrestricted commercial use, modification, and redistribution. The release includes both Base and Supervised Fine-Tuning (SFT) versions, with support already available for tools like ComfyUI.