OpenMOSS announced MOSS-Image-Base and MOSS-Image-SFT on October 16, with publicly available code and weights. The 5B-parameter image backbone targets text rendering, layout control and multiple visual styles.
The project describes prompts of up to 4K tokens and adaptation of the base model for custom tasks. It provides integrations for ComfyUI, SGLang and vLLM-Omni; actual resource needs depend on the deployment.
The useful release facts are its controllability and available artifacts. The roundup’s “no safety guardrails” framing is not treated here as a verified feature or as evidence that every request is supported.
Sources: Official repository and release notes · Official user guide