MOSI and the OpenMOSS team say their open image model, MOSS-Image, drew about 1.3 million downloads in four days. Pandaily reported the figure on 25 October, citing the team and QbitAI. It adds up several repositories, including Comfy-Org's on Hugging Face, so no single page shows it. Pandaily also notes about 1,400 GitHub stars and a top place in Hugging Face's text-to-image trends. It says none of this is independently audited.
Pandaily describes three parts. A frozen Qwen3.5-9B text encoder reads the prompt, a 5-billion-parameter diffusion transformer trained from scratch draws the image, and the FLUX.2 VAE converts it to pixels. For design work, the team offers JSON layout input with bounding boxes, which pins where elements sit, and prompts up to 4,096 tokens, which leaves room for detailed briefs. Ostris, a ComfyUI developer, suggested that leaving AI images out of training may explain the natural look. That is his view, not a test. Some testers found certain styles unstable. The licence is not stated.
A download total that sums several repositories measures curiosity, not adoption. Support in Diffusers, ComfyUI, SGLang and vLLM-Omni makes a trial cheap. Run a small test on your own designs and languages before reading the count as a market signal, and check the licence first.