ComfyUI-MOSS-Image-Studio-Lite-SatoDive

MOSS-Image Studio Lite - SatoDive (1.0.0)

Fast, simple MOSS-Image (OpenMOSS, 5B DiT + Qwen3.5-9B) in ComfyUI: one main node next to the native loaders, every setting a visible widget, a drawable Layout canvas, inpaint, and a Prompt Enhancer that uses the model's own text encoder.

Video Tutorial & Walkthrough

<p align="center"> <a href="https://www.youtube.com/watch?v=AWNb8O1pHVQ"> <img src="https://img.youtube.com/vi/AWNb8O1pHVQ/maxresdefault.jpg" alt="MiniMax-H3 ComfyUI Tutorial" width="750"> </a> </p> <p align="center"> ▶️ <i>Click the image above to watch the complete step-by-step walkthrough on YouTube.</i> </p>

📥 ## Download ComfyUI Workflow For Free: <a href="https://www.patreon.com/SatoDive/posts/moss-image-might-173268998">Download the Workflow</a>

Nodes

MOSS Generate Lite (SatoDive) - the whole job in one node

• Text-to-image with turbo defaults (8 steps, CFG 1 with the 8-step DMD LoRA).
• Layout: drag boxes on the node's canvas - move, resize, set each box's type (object / character / text), its own prompt and a strict strength (0 = free ... 1 = exactly your words, nothing else inside the box). Compiled to the official MOSS-Image Layout prompt.
• Inpaint / img2img: connect Load Image (paint the mask in the Mask Editor); only the painted area changes, the rest is pasted back pixel for pixel.
• Negative prompt and the official skip-layer guidance (layer 22) when CFG > 1.
• Saves to output/MOSS with the workflow inside the PNG; shows the last render's report and the exact final prompt.

✨ Enhance (button in the node) and MOSS Prompt Enhancer Lite (SatoDive)

• MOSS's own Qwen3.5-9B rewrites your plain words following the official MOSS-Image prompt guide: a detailed caption or a Layout (boxes) - with boxes, each box is rewritten on its own (strict 1 = never touched).
• Runs only when you click (a separate small job, nothing rendered); review Before / After, edit, then Apply.
• The node version is OFF by default (OFF = your text as it is, the LLM never runs).

The panel grows with its content and never shrinks a size you chose (right-click: Fit the panel now).

Install

1. ComfyUI v0.42 or newer (native MOSS-Image support).
2. Copy this folder into ComfyUI/custom_nodes and restart ComfyUI. No extra Python packages.
3. Models (Comfy-Org/MOSS-Image on Hugging Face):
   • diffusion_models: moss_image_sft_int8_convrot (or bf16)
   • text_encoders: qwen3.5_9b_int8_convrot (CLIPLoader type moss_image)
   • vae: flux2-vae
   • optional turbo LoRA: MOSS-Image-SFT-8-steps-dmd (lodestones) with LoraLoaderModelOnly at 1.0
4. Graph: UNETLoader -> LoraLoaderModelOnly -> MOSS Generate Lite, plus CLIPLoader and VAELoader.

Example workflows are free on SatoDive's Patreon.

Settings

| | steps | CFG | negative | SLG | |---|---|---|---|---| | Turbo (8-step DMD LoRA) | 8 | 1 | not used | off | | Full quality (no LoRA) | 25-50 | 2.5-3 | official default | on |

Changelog

• 1.0.0 - first release: MOSS Generate Lite (text-to-image, Layout canvas with strict boxes, inpaint), Enhance button + MOSS Prompt Enhancer Lite (caption / Layout / each box on its own), auto-grow panel.

License

MIT - SatoDive