Skip to main content
MOSS-Image is an open-source text-to-image model from the OpenMOSS team, released under the Apache 2.0 license. It follows long, detailed prompts and renders English and Chinese text, and Mage runs it on its own GPUs, so there is nothing to download. It runs unlimited from Basic at 1K and from Pro at 2K, and takes LoRAs and negative prompts.

​
Overview

​
Unlimited generation by plan

Which plan runs each resolution unlimited, at the model’s base settings. Gems means that setting always costs Gems on that plan. Add-ons such as LoRAs may raise a run’s price and move it to a higher plan or to Gems.

​
What it supports on Mage

​
Best for

  • Long, detailed prompts that describe a whole scene
  • Signs, posters, and menus with English or Chinese text
  • Styles from community MOSS-Image LoRAs

​
Using LoRAs

Pick a community MOSS-Image LoRA under Model → LoRA, or bring your own from CivitAI on the Import page: paste the model’s link and Mage recognizes the “MOSS Image” base model. Standard LoRA and LoCon files work; LoKr, LoHa, and DoRA files are not supported yet. See LoRAs for how weights and stacking work.

​
An example prompt to start from

​
Limitations

  • No image input, so no Characters, References, or edits.
  • It runs 50 steps at full quality, so it is slower than few-step models.

​
Use via API

MOSS-Image is available through the Mage API as the moss_image architecture. Its fields, tokens, defaults, and price are on its API reference page.
Last modified on October 24, 2026