Feedback
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
MOSS-Image
MOSS-Image lets you steer composition with structured prompts: describe a scene in JSON and watch a 5B model render each element in the spot you assigned.
AI Tools
Explore AI tools
Why Creators Pick MOSS-Image
An open 5B image family where the prompt itself decides placement, not chance.
- Placement Set by PromptHand each object and text block its own coordinate box, so the finished composition mirrors your spec rather than an approximation.
- 5B Model, 4K-Token ContextLengthy briefs stay in memory without being cut short, and the released weights can be fine-tuned for your own niche.
- Base + SFT CheckpointsTwo Apache-2.0 releases share one architecture; the fine-tuned build keeps whiteboard captions cleaner and better grouped.
Running MOSS-Image in Three Moves
From installing dependencies to your first coordinate-driven render.
MOSS-Image Questions, Answered
Prompting, checkpoints, hardware and licensing — the details people ask about most.
What can I control with a layout prompt?
Your JSON names the scene, the objects and any rendered text, then gives each one a box that sets where it sits and how large it appears.
How is the SFT checkpoint different from Base?
Both share one architecture and sampling API. SFT was trained on curated data, and in the project demo it groups text more clearly and produces a neater whiteboard.
Is there a limit on prompt length?
Roughly 4K tokens — plenty of room for a multi-element layout written out as JSON.
Which hardware and software are required?
Python 3.10+ alongside an NVIDIA GPU and a matching PyTorch build. Low on VRAM? Switch to Sequential mode or shrink the canvas; Compile DiT stays off by default and is unavailable there.
Does it work with ComfyUI?
It does. Install the native plugin through ComfyUI Manager, then open the sample workflow; merging into the official ComfyUI repository is still underway.
What licensing covers the weights and code?
The weights ship under Apache-2.0, while the pipeline code and bundled third-party parts keep their own terms. Cite the project repository in research write-ups.
Render Your First Layout with MOSS-Image
Set up the Diffusers pipeline or the ComfyUI plugin, then feed it a structured layout — or let the Prompt Enhancer shape your idea before you render.