MOSS-Image and its versions
MOSS-Image is the open image model from the OpenMOSS team, built for accurate Chinese and English text and layouts you can place with JSON. It was released on 16 October 2026 as two checkpoints; fal hosts the fine-tuned one, which has its own page with the same twelve prompts as every image model here.
OpenMOSS · facts checked
Which one to pick
Side by side, measured
| Model | Price | Seconds per image | 12 prompts cost | Sizes |
|---|---|---|---|---|
| MOSS-Image | $0.05 per megapixel via fal | 14 to 63 | $0.6979 | Any width and height from 512 to 2048 pixels, rounded to multiples of 16 |
Seconds and cost are from our own runs of the same 12 prompts through each model; open a model for every picture and the request behind it.
Every version
| Version | Released | What changed |
|---|---|---|
| MOSS-Image-Base MOSS-Image-Base | The 5 billion parameter base text-to-image model, for fine-tuning. | |
| MOSS-Image-SFT fal-ai/moss-image | Fine-tuned from the base on curated data; the version on fal, at $0.05 a megapixel, with a layout planner that writes a JSON plan from your prompt. |
Frequently asked questions
Is MOSS-Image open source?
The pipeline code is Apache 2.0 and both checkpoints are downloadable from OpenMOSS. Check the model cards for the weights' licence before commercial self-hosting.
Who makes MOSS-Image?
The OpenMOSS team, which published the base and SFT checkpoints and the inference pipeline on GitHub.