MOSS-Image Generator — Layout-Aware Text-to-Image
freeTrialImage.bannerPity

Feedback

Result
Preview

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

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

MOSS-Image

Try MOSS-Image online free — a 5B text-to-image model that reads 4K-token prompts and JSON layouts to place objects, text and boxes exactly where you want.

AI Tools

Explore AI tools

placeholder hero

Why Creators Choose MOSS-Image

An open 5B image model from the MOSS-Image family that treats your layout instructions as a blueprint, not a suggestion.

  • Layout Commands, Not Guesswork
    Assign every object and caption its own bounding box, and the frame is composed the way your JSON file dictates.
  • Compact 5B Core, Long Context
    Detailed briefs stay intact inside a 4K-token window, and the open weights are straightforward to fine-tune for niche domains.
  • Two Checkpoints, One License
    Base and SFT builds share the same architecture under Apache-2.0; the tuned version keeps labels cleaner on whiteboard-style scenes.

Running MOSS-Image in Three Moves

From installing dependencies to a fully controlled MOSS-Image layout render.

FAQ

Questions About MOSS-Image

Straight answers on MOSS-Image prompts, checkpoints, hardware and licensing.

1

What exactly can I control with a layout prompt?

Each JSON entry defines the scene, the objects inside it and any rendered text, plus a bounding box that fixes position and size.

2

What separates MOSS-Image-Base from the SFT build?

Both use one architecture and the same sampling interface. SFT has been fine-tuned on curated data and, in the project demo, groups text more clearly on whiteboard scenes.

3

Is there a limit on prompt length?

You get a 4K-token window — plenty of room for a multi-element layout file.

4

Which hardware and software are required?

Python 3.10+ plus an NVIDIA GPU and a matching PyTorch build. Short on VRAM? Move to Sequential mode or shrink the canvas — Compile DiT is disabled by default and cannot run in that mode.

5

Does it work inside ComfyUI?

Yes — install the native plugin via ComfyUI Manager, then open the bundled workflow. A merge into the official ComfyUI repository is still underway.

6

Which license covers the weights and the code?

Weights ship under Apache-2.0; the pipeline code and third-party parts retain their own terms. Cite the repository in research work.

Put MOSS-Image to Work

Spin up the Diffusers pipeline or the ComfyUI plugin, feed MOSS-Image a layout prompt — and let the Prompt Enhancer sharpen your idea first if you prefer.