OpenMOSS releases MOSS-Image: 5B open text-to-image model with JSON layout control
AdinaYakup · x · 2026-10-17
The OpenMOSS team open-sourced MOSS-Image-Base, a 5B text-to-image model under Apache 2.0. It supports natural-language prompts, tag sequences, and JSON layout control for text placement and object arrangement, with up to 4K token prompts. The model card demos fine Chinese handwritten text rendering; Base and SFT versions are provided (SFT gives clearer text grouping and tidier whiteboard layouts). Available on Hugging Face via a custom Diffusers pipeline, with ComfyUI support.
Related event: Fudan's OpenMOSS Open-Sources 5B MOSS-Image Text-to-Image Models(14 posts)→
More from Multimodal
- Leak: xAI building Director, an AI film studio inside Grok Imagine with 24 AI voices — nima_owji · 2026-10-20
- Creator rebuilds an Abu Simbel documentary with Codex scripting and ElevenLabs voiceover — aziz4ai · 2026-10-20
- Midjourney --sref combo generates fuzzy textured blue animal characters — sergeantsref · 2026-10-20
- 768p video with audio generated on just two desktop DGX Sparks using new FastVideo QAD-v2 — Scobleizer · 2026-10-20
- MOSS image SFT at just 6 steps shows strong detail, beating local Qwen in tests — DevKkw · 2026-10-20
- AI-Generated Music Video 'Mike from Jersey' Shows Cartoon Deli Scene — Ass_Blast69 · 2026-10-20