MOSS Image AI

What is MOSS-Image?

MOSS-Image is a 5-billion-parameter text-to-image model family published by OpenMOSS. The project releases model weights, inference examples, and ComfyUI workflows for local use.

Mossimage.run is an independent studio built around that model family. It is not the OpenMOSS project website and is not affiliated with OpenMOSS or fal.ai.

Base and SFT serve different workflows

Base is the foundation checkpoint intended for evaluation, LoRA training, and custom post-training. SFT is the supervised fine-tuned checkpoint intended for more direct image generation. Both are released under Apache 2.0.

This online studio uses the hosted SFT endpoint. Base and SFT can both be run locally with their matching model files and ComfyUI workflows.

Request English and Chinese text inside an image

Prompts can include quoted wording for posters, menus, labels, and visual concepts. MOSS-Image supports English, Chinese, and bilingual requests, although every generated word still needs proofreading before publication.

Guide composition with structured JSON

JSON Layout prompts describe the overall scene and list intended regions for text or objects. MOSS-Image treats those bounding boxes as experimental layout guidance rather than exact design coordinates.

Try SFT online or run the checkpoints locally

Use the browser studio when you want to test prompts without downloading weights. Use the local path when you need Base, a custom ComfyUI graph, or control over model files and hardware.