Repolode
Submit
OpenMOSS

OpenMOSS/MOSS-Image

Open-source 5B-parameter text-to-image diffusion transformer with strong in-image text rendering and JSON layout control.

Open-source alternative to:MidjourneyAdobe Firefly

View on GitHubWebsite
Stars
647
Forks
26
Open issues
6
Last commit
2 days ago

What is it?

MOSS-Image is a text-to-image model from OpenMOSS, released as a base model (MOSS-Image-Base) and a fine-tuned variant (MOSS-Image-SFT). It uses a 5B-parameter diffusion transformer and is aimed at prompts where text must appear accurately inside the picture, such as posters, signs and handwritten notes.

Beyond plain prompts, the model accepts structured JSON layouts that place text and objects at specified bounding boxes on a canvas. Prompts of up to 4K tokens are supported, and the project includes a Python pipeline usable through Diffusers, a Gradio web app, and a ComfyUI plugin. Memory modes from fully GPU-resident to sequential offloading are available for constrained hardware.

The repository also ships a prompt enhancer skill that converts image ideas into detailed captions or layout JSON, and documentation covers LoRA fine-tuning for custom styles and domains.

Key features

  • 01Generates images with clear, accurate text rendering.
  • 02Controls text and object placement with JSON layouts using bounding boxes.
  • 03Accepts prompts up to 4K tokens long.
  • 04Provides Python, Gradio and ComfyUI interfaces.
  • 05Offers memory modes for different GPU capacities.
  • 06Supports LoRA fine-tuning for custom tasks.

Use cases

  • Creating posters or banners where exact title and body text must appear in the image.
  • Generating mock-ups such as signage, sticky notes or whiteboards with readable text.
  • Producing layout-controlled compositions for marketing or design drafts.
  • Fine-tuning a base image model for a specific art style or domain.

Topics

#text-to-image#diffusion-model#image-generation#text-rendering#layout-control#fine-tuning#comfyui#controllable-generation#foundation-models

FAQ

What is MOSS-Image?
Open-source 5B-parameter text-to-image diffusion transformer with strong in-image text rendering and JSON layout control. MOSS-Image is a text-to-image model from OpenMOSS, released as a base model (MOSS-Image-Base) and a fine-tuned variant (MOSS-Image-SFT). It uses a 5B-parameter diffusion transformer and is aimed at prompts where text must appear accurately inside the picture, such as posters, signs and handwritten notes.
Is MOSS-Image free to use? What license does it have?
MOSS-Image is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is MOSS-Image still maintained?
Actively maintained. The last commit to MOSS-Image was on Oct 16, 2026.
What can I use MOSS-Image for?
Creating posters or banners where exact title and body text must appear in the image. Generating mock-ups such as signage, sticky notes or whiteboards with readable text. Producing layout-controlled compositions for marketing or design drafts. Fine-tuning a base image model for a specific art style or domain.
Where can I get MOSS-Image?
The source code and installation instructions are on GitHub: https://github.com/OpenMOSS/MOSS-Image

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects