Qwen-Image-2512: More Refined Details, Higher Realism

Qwen-Image-2512 is a December update to the Qwen-Image base model released in August, significantly reducing the AI feel in human representation and boosting overall realism. Details such as natural landscapes and animal fur have also become more refined, and the accuracy and layout quality of text rendering have improved.

It performed strongly in over 10,000 AI Arena blind evaluations, and is now introduced as a powerful open-source model. In particular, for portrait-photo-style results, facial detail, strand-by-strand hair representation, and age reproduction are much more natural than in the previous version.

The improvements can be summarized into three main points.

• Human realism: More realistic representation not only of young subjects but also wrinkles, skin texture, and expressions of elderly subjects
• Natural detail: Enhanced fine texture and material representation for rivers, waterfalls, waves, forests, and animal fur
• Text rendering: Improved complex text placement and readability, such as PPT slides, infographics, and text within images

The examples highlight these changes through scenes such as a selfie with a dormitory background, a person at an anime event venue, a student standing on a school playground, an elderly couple in a kitchen, canyon and ocean landscapes, and a golden retriever and an argali. Finally, the Qwen team requests that the Qwen-Image Technical Report be cited when used in research or applications.