Alibaba Unveils Next-Gen Image Generation AI 'Qwen-Image 3.0'

Chinese Alibaba unveiled its next-generation image generation AI 'Qwen-Image 3.0' on the 22nd (local time). Alibaba Cloud stated that the core keyword of this model is 'Real,' revealing that it was developed with the goal of creating task-optimized AI that enhances practicality and realism, moving beyond the stage of simply generating aesthetic images.

The maximum prompt length the model can process at once has been expanded to 4,500 tokens. Consequently, it can implement layouts containing large amounts of information, such as 3x3 composite infographics or composite user interfaces (UIs) using a 'Picture-in-Picture' method, as a single image. By applying ultra-high-precision expression capabilities, it achieves high resolution that can accurately read even small text as tiny as 10 pixels (px). It supports 'Authentic Details' by realistically reproducing fine details such as LaTeX formulas, newspaper layouts, and skin pores, as well as restoring damaged artifacts.

It supports 12 languages, including Korean, by default and expresses over 100 types of art styles and various fonts. Based on this, it can naturally generate various digital user environments (UIs), such as web pages and game screens. Alibaba plans to expand its use in productivity-centric areas such as education, content production, and e-commerce through this model.