Moss Image

Moss Image puts MOSS-Image, the open text-to-image model from the OpenMOSS team, in a browser tab, so you can test prompts without a GPU, Docker image or local install. The model is a 5B-parameter diffusion transformer that reads prompts through a Qwen3.5 text encoder and decodes pixels with the FLUX.2 VAE, which keeps skin, fabric and small type sharp. Features: text to image from a plain-English prompt; readable English and Chinese lettering inside the image for posters, covers, banners and title cards; JSON layout prompts that place each element where you ask; Base (for LoRA / fine-tuning) and SFT checkpoints, the site runs SFT; advanced settings for steps, guidance, negative prompt, seed and HD/2K canvas. Free credits at sign-up; paid plans add commercial license, no watermark and priority queue. Independent site, not affiliated with OpenMOSS.