Alibaba's new open Qwen image model aims for more natural-looking results

Dec 31, 2025

Alibaba has released Qwen-Image-2512, an update to its text-to-image model. The company says the new version does a better job generating realistic images of people, with finer facial detail and less of the artificial look that plagued earlier versions.

The updated image model aims to eliminate the "plastic" look of its predecessor. | Image: Qwen

The display of text in images, for example, in infographics or presentations, has also been improved. Landscapes, animal fur, and other natural elements also come out with finer detail.

The new Qwen model handles text in images better than before. | Image: Qwen

In over 10,000 blind tests on Alibaba's AI Arena platform, Qwen-Image-2512 came in fourth overall, making it the top-ranked open-source model, according to Alibaba. It's up against other open models like HunyuanImage-3.0, Z-image, and Flux.2.

Qwen-Image-2512 is available on Hugging Face and ModelScope, and you can try it out through Qwen Chat. More details are in the Tech Report and on the blog.

AI News Without the Hype – Curated by Humans

As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.

Source: Hugging Face / Qwen-Image-2512 | ModelScope / Qwen-Image-2512 | Qwen Chat / Demo | Qwen / Tech Report | Qwen / Blog