Qwen-Image-2.1 文生图能力评测：Lighting、Knowledge、Reasoning 最接近前沿
Where Qwen-Image-2.1 is strongest in Text to Image: closest to the category frontier among all models in Lighting, Knowledge and Reasoning
Artificial Analysis 的 Text to Image 评测显示，Qwen-Image-2.1 在 Lighting 能力上最接近该类别前沿，其后依次是 Knowledge、Reasoning 和 Layout。相比 Qwen Image 2.0，它在全部能力上都缩小了与前沿的差距，其中 Layout、Lighting 和 Knowledge 提升最大。该评测体系覆盖 9 项能力和 10 类用例，每项各有独立榜单。
正文
当前显示原文，中文全文译文尚未就绪。
Where Qwen-Image-2.1 is strongest in Text to Image: closest to the category frontier among all models in Lighting, Knowledge and Reasoning
Our Text to Image taxonomy measures 9 capabilities and 10 use cases, each with its own leaderboard. Capabilities are the individual skills that go into making an image.
Qwen-Image-2.1 is closest to the category frontier in Lighting, followed by Knowledge, Reasoning and Layout. Against Qwen Image 2.0 it closes the gap to the frontier on every capability, with the largest gains in Layout, Lighting and Knowledge.
➤ Lighting covers reflection, refraction, shadows, and caustics.
➤ Knowledge covers real landmarks, species, and domain facts across science and common sense.
➤ Reasoning covers idiom interpretation, maths and science, spatial and temporal reasoning, and concept mixing.