MOSS-Image Generator: Turn Words or JSON Blueprints into Art
freeTrialImage.bannerPity

Feedback

Result
Preview

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

MOSS-Image

Design precise AI images with MOSS-Image: a five-billion-parameter model that follows text or JSON blueprints to place every element.

AI Tools

Explore AI tools

placeholder hero

Why Creators Choose MOSS-Image

An open image model family that treats composition as part of the prompt, not an afterthought.

  • Layout Written Into the Prompt
    Every object and caption carries its own frame, so the composition follows your blueprint rather than a lucky guess.
  • Five Billion Parameters, 4K Context
    Detailed briefs fit without trimming, and the released weights can be fine-tuned for your own niche or workflow.
  • Two Checkpoints, One Architecture
    Base and SFT builds share the same design and ship under Apache-2.0, while the tuned version groups whiteboard lettering more neatly.

Running MOSS-Image in Three Moves

From installing dependencies to exporting a precisely placed composition.

FAQ

Questions About MOSS-Image

Answers on prompt length, checkpoints, hardware needs and licensing.

1

What exactly can I position with a layout prompt?

Your JSON names the scene, the objects and any lettering, then assigns a frame with coordinates and dimensions to each one.

2

How is the SFT checkpoint different from Base?

Both use the same design and sampling calls, but SFT was trained further on hand-picked samples, which in the demo yields cleaner text grouping and a neater whiteboard scene.

3

Is there a limit on prompt length?

Roughly 4,000 tokens, which comfortably covers a multi-element blueprint.

4

Which GPU and Python version should I use?

You need Python 3.10 or later plus an NVIDIA card running a compatible PyTorch build. When VRAM is limited, move to Sequential mode or shrink the canvas; Compile DiT stays disabled by default and cannot run in that mode.

5

Does it work as a ComfyUI node?

It does — install the native plugin through ComfyUI Manager, load the bundled workflow and start rendering. Listing in the official ComfyUI repository is still pending.

6

Under what license are the weights released?

The weights ship under Apache-2.0; the pipeline code and bundled third-party parts follow their own terms. If you publish research, please cite the project repository.

Try MOSS-Image on Your Next Composition

Set up the Diffusers pipeline or the ComfyUI plugin, drop in a blueprint prompt, or let the built-in prompt helper shape your idea before you render.