MOSS-Image SFT: 5B Text-to-Image Model with Layout Control
MOSS-Image SFT is the supervised fine-tune of OpenMOSS' 5B image model, with layout JSON control, English and Chinese text rendering, and ComfyUI support.
MOSS-Image SFT
Text-to-ImageLayout ControlText RenderingOpen WeightsThe supervised fine-tune of MOSS-Image on curated data, and the checkpoint to run for text-to-image work. It keeps the same architecture and pipeline as the base model while producing tidier text grouping and more organized layouts, and it supports JSON layouts that place text and objects by bounding box.
| Developer | OpenMOSS |
| Release Date | 2026-10-16 |
| Architecture | 5B single-stream DiT (42 layers) in the FLUX.2 latent space, Qwen3.5-9B text encoder, FLUX.2 VAE |
| Parameters | 5B |
| License | Apache-2.0 |
| Sampling | 50 steps or more, CFG 3, shift 5, Euler, negative-branch skipping at layer 22 |
Capabilities
- Text rendering: legible English and Chinese text inside the image, with automatic line breaks
- Layout JSON: a
summaryplusitemswith bounding boxes controls where text and objects land, and what the text says - Prompt formats: natural language, booru-style tag sequences or layout JSON, with prompts up to 4K tokens
- Resolution range: 1M, 1.5M and 2M pixels recommended, 0.25M to 4M supported, multiples of 16
- Style range: both anime and photorealistic output, according to OpenMOSS
Notes
MOSS-Image-SFT is fine-tuned from MOSS-Image Base and is the checkpoint most ComfyUI users will run; the repackaged weights ship both checkpoints. Sampling settings from the project repository are 50 steps, CFG 3, shift 5 and negative-branch skipping at layer 22 with Euler and sgm_uniform.
Related
Guides and workflows related to this model series.
Comments
Sign in with GitHub to join the discussion.