OpenMOSS-Team/MOSS-Image-SFT · Hugging Face

Source: https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT

MOSS-Image-SFT is a supervised fine-tuned text-to-image model in theMOSS-Imagefamily from OpenMOSS, derived fromMOSS-Image-Base. It uses the same architecture and inference pipeline, supporting natural-language prompts, tag sequences, and JSON layouts for controlling text placement and object arrangement.

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#base-vs-sftBase vs. SFT

In this example, SFT produces clearer text grouping and a more organized whiteboard layout.

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#quick-startQuick start

Use Python 3.10 or later and an NVIDIA GPU with CUDA support. Install a PyTorch 2.9 or later build compatible with your CUDA environment, then install:

pip install "diffusers>=0.38" "transformers>=5.3" "accelerate>=1.10" "safetensors>=0.6" "Pillow>=10"

import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "OpenMOSS-Team/MOSS-Image-SFT",
    custom_pipeline="pipeline_moss_image",
    trust_remote_code=True,
    torch_dtype={"default": torch.bfloat16, "vae": torch.float32},
)
pipe.scheduler = type(pipe.scheduler).from_config(pipe.scheduler.config)
pipe.to("cuda")

prompt = '''A wide view of a refrigerator beside a kitchen counter with a fruit bowl. A small sticky note on the refrigerator door is handwritten in Chinese:
"记得买牛奶和鸡蛋！
牛奶两盒，鸡蛋一打。
顺便带一袋面包。"'''

image = pipe(
    prompt,
    height=1024,
    width=1024,
    num_inference_steps=50,
    guidance_scale=3.0,
    uncond_skip_layers=[22],
    generator=torch.Generator("cpu").manual_seed(42),
).images[0]
image.save("refrigerator_note.png")

The model and custom pipeline code are downloaded and cached automatically. You can also pass a complete local model directory tofrom\_pretrained. The transformer and text encoder use bfloat16; the VAE and initial noise use float32. The bundled VAE uses the FLUX.2 small decoder.

For the Gradio interface, CPU offload, and compilation options, follow theproject quick start. For ComfyUI, search forMOSS-ImageinComfyUI Managerto install the native plugin. Integration into the official ComfyUI repository is in progress. See theinstallation and usage guideandexample workflow.

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#sampling-settingsSampling settings

SettingDefaultSamplerFlow-matching EulerSteps50CFG3Shift5Layer skippingNegative branch, layer index 22 (zero-based)Negative promptEmpty The default guidance uses two transformer forwards per step. Passuncond\_skip\_layers=\[\]to disable negative-branch skipping, or setnegative\_promptto provide a negative prompt. Python API image dimensions must be positive multiples of 16.

See theuser guidefor resolution, sampling, prompt, and LoRA training recommendations.

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#prompt-enhancer-skillPrompt enhancer skill

TheMOSS-Image Prompt Enhancerturns image ideas into detailed captions or spatial Layout prompts. It is designed for GPT-6-family agents with local file and Python tools. The bundled script converts pixel-coordinate drafts into compact JSON with integer centered coordinates; it does not call an image-generation service.

Install theskill folderin your agent’s skills directory, then invoke$moss\-image\-prompt\-enhancerwith your image idea, desired caption or Layout output, and any canvas dimensions. Paste the compiled result into MOSS-Image’s positive prompt; for Layout, use the same generation dimensions.

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#layout-to-image-generationLayout-to-image generation

Pass a JSON layout as the prompt to describe the scene, rendered text, objects, and their positions. Layouts usesummary\.captionanditems, withbbox\_format: "yxyx"andbbox\_coord\_space: "abs\_centralized".

With the Python API, serialize the layout usingjson\.dumps\(layout, ensure\_ascii=False\)and pass it topipe. In Gradio, paste the JSON into the Prompt field and set the matching image dimensions. See thelayout format and example.

Layouts can also support controlled editing and user-friendly layout tools. Edit text, element descriptions, and bounding boxes to adjust image content and composition in a structured way.

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#showcaseShowcase

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#citationCitation

@misc{mossimage2026,
  title = {MOSS-Image},
  author = {{OpenMOSS Team}},
  year = {2026},
  howpublished = {\url{https://github.com/OpenMOSS/MOSS-Image}}
}

https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT#license-and-acknowledgementsLicense and acknowledgements

The model weights and bundled inference code are released under theApache License, Version 2.0.

We thank the Diffusers, Qwen, and FLUX teams.