# MOSS-Image Base vs SFT: Which Checkpoint Should You Use?

The MOSS-Image release ships with more than one checkpoint, and picking the right one saves time. Here is a short comparison, plus a way to try the tuned model online through Moss Image. For the live version, visit [Moss Image](https://moss-image.net/).

## A simple poster workflow

Start with a short prompt that names the subject, the style and the exact headline in quotes. Generate a few images with different seeds, keep the best seed, then refine the wording or move elements with a JSON layout prompt. Finish by switching to the HD or 2K canvas for the final render.

## Tips for better lettering

Keep headlines short, put the exact words in quotes and say which language you want. Ask for one text block at a time if the layout is busy. If a word comes out wrong, keep the seed and adjust guidance slightly within the 3.0 to 6.0 range instead of rewriting the whole prompt.

## Commercial use

If the images are for client work or products, check the plan you are on. Paid plans add a commercial license, remove the watermark and give you a priority queue and private generation, which keeps unreleased work out of public view.

## Licensing and independence

The MOSS-Image model weights are released under Apache-2.0. The online site is an independent service that runs the model and is not affiliated with the OpenMOSS team. Questions about the site itself go to support@moss-image.net.

## What MOSS-Image is

MOSS-Image is an open text-to-image model released by the OpenMOSS team. It is a 5B-parameter diffusion transformer. Prompts are read through a Qwen3.5 text encoder, and pixels are decoded with the FLUX.2 VAE. That combination is a large part of why skin, fabric and small type come out sharp instead of smeared.

## Bottom line

Moss Image is worth a try if this matches what you need. Moss Image is an independent site and is not affiliated with the OpenMOSS team; the MOSS-Image model weights are released under Apache-2.0.