Alibaba publishes the Qwen-Image-2.0 technical report

The reported 7B image generation and editing model combines a Qwen3-VL encoder with a multimodal diffusion transformer. It produces native 2K output, accepts up to 1,000 prompt tokens, and improves multilingual text in slides, posters, infographics, and comics. Alibaba reported 0.91 on GenEval, 88.32 on DPG-Bench, and first place in both generation and editing on the cited AI Arena Elo rankings.

Visit original source