Just saw Qwen-Image pop up on Hugging Face, this text-to-image model’s been getting some buzz lately. The model page tags have a decent amount of info: supports Chinese and English (en/zh), runs on QwenImagePipeline, works directly with diffusers, license is apache-2.0, and it also mentions Azure deployment plus an arxiv ID.
For someone like me who wants to integrate it into a project, that apache-2.0 part is the most critical—commercial-friendly, saves a ton of hassle with legal stuff. The native diffusers pipeline also means less headache setting it up. Native Chinese support is a big plus for domestic products too, since a lot of overseas models struggle with understanding Chinese prompts.
Haven’t done a full-scale run on image quality yet, just laying out the info here. Anyone who’s already tried it, how’s the actual performance, especially with long Chinese prompts—does it keep up?