MOSS-Image Generator: Layout-Aware Image Creation
freeTrialImage.bannerPity

Feedback

Result
Preview

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

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

MOSS-Image

Explore MOSS-Image: an open text-to-image model that reads JSON layouts and 4K-token prompts to place every object, label and box exactly where you want.

AI Tools

Explore AI tools

placeholder hero

Precision Layout Rendering with MOSS-Image

An open 5B image family from the MOSS-Image project, where the prompt doubles as a blueprint for composition.

  • Layout Built Into the Model
    Assign a box to every object and caption, so the composition obeys your JSON instead of leaving placement to chance.
  • Five-Billion-Parameter Core, 4K Context
    Long, detailed briefs stay in memory, and the same weights can be tuned for your own domains and tasks.
  • Two Checkpoints, One Architecture
    Base and SFT variants share identical structure under Apache-2.0; the fine-tuned build keeps whiteboard lettering cleaner.

Get MOSS-Image Running: Step by Step

From installing dependencies to a layout you fully control — three short moves.

FAQ

MOSS-Image: Your Questions Answered

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

1

What exactly can I control with a layout prompt?

With MOSS-Image, your JSON describes the scene, its objects and any rendered text, plus a bounding box for each element's position and size.

2

How is the SFT checkpoint different from Base?

Both share one architecture and sampling interface. SFT is fine-tuned on curated data, which in the project demo yields tighter text grouping and a cleaner whiteboard result.

3

How much text can one prompt hold?

MOSS-Image accepts up to roughly 4K tokens — plenty of room for a multi-element layout written as JSON.

4

Which hardware and software are required?

MOSS-Image needs Python 3.10 or newer plus an NVIDIA GPU with a matching PyTorch build. Short on VRAM? Switch to Sequential mode or lower the image size; Compile DiT is off by default and unsupported there.

5

Does it work inside ComfyUI?

Yes — find the native plugin through ComfyUI Manager, install it, then load the sample workflow. Listing in the official ComfyUI repository is still underway.

6

What licensing covers the weights and code?

Weights ship under Apache-2.0; the pipeline code and third-party components keep their own terms. Cite the project repository for research use.

Put MOSS-Image to Work Today

Set up the Diffusers pipeline or the ComfyUI plugin, then test a JSON layout prompt — or shape your idea with the MOSS-Image prompt helper before you render.