Feedback
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
MOSS-Image
Try MOSS-Image online: a 5B text-to-image model that follows JSON layouts and keeps long prompts, so objects, labels and boxes land exactly where you plan.
AI Tools
Explore AI tools
Precision by Design: Inside the MOSS-Image Model
A text-to-image model line that reads long prompts and honors structured layouts.
- Layout Directives, Not GuessworkEach object, caption and bounding box gets its own coordinates, so the render follows your JSON map instead of making assumptions.
- Long-Context 5B BackbonePrompts up to 4K tokens stay readable, and the same weights can be tuned for your own domains and tasks.
- Base and SFT CheckpointsBoth variants share one architecture under Apache-2.0; the SFT release is tuned for cleaner text grouping and tidier whiteboard scenes.
Run MOSS-Image in Three Steps
From Python setup to a layout-controlled render, follow this short workflow.
MOSS-Image Questions, Answered
Answers about layout prompts, model checkpoints, hardware needs and ComfyUI setup.
What can I define with native layout control?
Your JSON prompt can name the scene, objects and visible text, then assign a position and size box to each element.
How do Base and SFT checkpoints compare?
They use the same architecture and sampling interface; SFT is fine-tuned on curated data and, in the project demo, produces clearer text grouping and a neater whiteboard layout.
What is the maximum prompt length?
Prompts can reach 4K tokens, which is enough room for a detailed multi-element layout JSON.
Which hardware and software are required?
You need Python 3.10+ and an NVIDIA GPU with a matching PyTorch build. If VRAM is limited, switch to Sequential mode or lower the image size; Compile DiT is off by default and unavailable in that mode.
Does MOSS-Image work with ComfyUI?
Yes. Install the native plugin from ComfyUI Manager, then open the example workflow; official ComfyUI repository integration is still underway.
What licensing applies to weights and code?
The model weights use Apache-2.0, while pipeline code and third-party components keep their own licenses. For research, cite the project repository.
Create Your First MOSS-Image Layout
Install the Diffusers pipeline or ComfyUI plugin, then test a JSON layout prompt — or shape your idea with the MOSS-Image Prompt Enhancer first.