Report
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
Comfyui Moss Image
Build precise scenes with Comfyui Moss Image: a ComfyUI node that reads JSON layout prompts and bounding boxes and renders crisp text with Base or SFT weights.
AI Tools
Explore AI tools

Meet Comfyui Moss Image: A Compact Layout-First Model
A small-but-capable 5B image family that places objects via structured JSON boxes, keeps on-image lettering legible, and handles both anime and photoreal looks.
- Structured Boxes Instead of GuessworkEvery entry in a JSON prompt carries its own box, so object placement, caption size, and scene composition follow your plan rather than the model's instinct.
- Long Prompts on a Light BackboneFeed in up to 4K tokens of scene detail while the 5B core keeps memory modest; Base and SFT weights plug into the same sampling setup.
- Readable Lettering Across Art StylesCaptions and labels stay sharp whether the output leans illustration or photography, covering posters, whiteboards, and product mockups alike.
From Install to Render with Comfyui Moss Image
Three short steps take you from a fresh plugin install to a polished, box-aligned image.
Questions About the Comfyui Moss Image Node
Answers on setup, checkpoints, box-driven prompts, and memory tuning inside ComfyUI.
Is a core-repository release required before I can use it?
Not at all — the community plugin is already searchable through ComfyUI Manager; a merge into the official repository simply is not finished yet.
How does the SFT checkpoint differ from Base?
They share one architecture and the same sampling controls; in the reference example the SFT weights produced tighter text grouping and a cleaner whiteboard composition.
What role do bounding boxes play in a prompt?
Each box pins one element's position and footprint, so editing coordinates, labels, or descriptions reshuffles the whole composition on the next run.
How long can prompts be, and which styles work best?
Up to 4K tokens per prompt, with anime and photoreal results demonstrated across captions, illustrations, photography, and design layouts.
My GPU has limited VRAM — which mode should I pick?
Advanced settings expose GPU-resident, Component, and Sequential loading; drop to Sequential or shrink the canvas when memory is tight, and note that Compile DiT ships off by default and is unavailable in Sequential.
Are the weights open for fine-tuning?
Yes — the 5B backbone can be tuned for niche domains, and the released weights carry an Apache-2.0 license, while pipeline code and third-party parts keep their own terms.
Start Rendering with Comfyui Moss Image
Grab the node from ComfyUI Manager, load the sample workflow, and push your first box-driven prompt through Base or SFT weights.




