OpenMOSS releases MOSS-Image: open-source 5B text-to-image model with native JSON layout control
lmoroney · x · 2026-10-18
OpenMOSS released MOSS-Image (Oct 16, Apache 2.0), a text-to-image model family with a Base model and an SFT version fine-tuned on curated data.
Its standout feature is native layout control: beyond natural-language prompts (up to 4K tokens), you can pass a structured JSON prompt specifying objects, rendered text, and their spatial positions via bounding boxes, with support for controlled layout-based editing.
Other features: compact 5B backbone, strong text rendering, anime and photorealistic styles, training on heterogeneous GPU platforms. Python/Gradio interfaces share a Diffusers pipeline; a live demo, ModelScope weights, and a technical report are available.
Related event: Fudan's OpenMOSS Open-Sources 5B MOSS-Image Text-to-Image Models(14 posts)→
More from Multimodal
- 'The Glitch': A Robot Folk Horror Short With a Cracked-Doll Android — BotLordBotflix · 2026-10-19
- tscomfy: Type-Safe ComfyUI Workflows as Code, Types Generated From Your Server — No_One_9304 · 2026-10-19
- 'Connect' Short Film Shows AI Video Closing In on Photoreal Human Eyes — indiegameplus · 2026-10-19
- MOSS image model design test: consistent pig character cards across six actions — DevKkw · 2026-10-19
- French #slopcore AI track went viral without lyrics, now gets an English-subtitled version — banteg · 2026-10-19
- AI-Generated Claymation-Style Horror Short 'Satanic Verse' Turns Heads on Reddit — Pnarpok · 2026-10-19