阿里巴巴Qwen-Image-2.1成开源图像模型第一
阿里新发布的Qwen-Image-2.1在图像生成和编辑基准上超越所有开源模型，支持原生2K输出和透明图像生成。
阿里巴巴于9月20日发布Qwen-Image-2.1，拥有70亿参数的视觉生成组件。该模型在AA-Image-T2I v2.0和AA-Image-Editing v2.0基准测试中排名第18位，超越Ideogram 4.0和HunyuanImage 3.0 Instruct，成为所有开源模型中的第一名。相比前代Qwen Image 2.0，排名从第72位和第58位大幅提升至第18位。
Alibaba's Qwen-Image-2.1 is the new #1 open weights model on both the AA-Image leaderboards
Qwen-Image-2.1 is Alibaba's latest image model, released with open weights on September 20. It has a 7B parameter visual generation component and handles Text to Image and Image Editing in a single model, with native 2K output and native transparent (RGBA) image generation and editing. The weights are released under the Qwen Research License, which requires a separate license for commercial use.
In the absence of major inference providers hosting this model at time of launch, we evaluated Qwen-Image-2.1 by hosting it locally.
Qwen-Image-2.1 ranks #18 on AA-Image-T2I v2.0 and on AA-Image-Editing v2.0, and leads all open weights models on both, ahead of Ideogram 4.0 (Quality) in Text to Image and HunyuanImage 3.0 Instruct in Image Editing. It is a large step up on Qwen Image 2.0, climbing from #72 to #18 in AA-Image-T2I v2.0 and from #58 to #18 in AA-Image-Editing v2.0.
The weights are available on Hugging Face and ModelScope, with day-0 support in Diffusers, ComfyUI, vLLM-Omni, SGLang and LightX2V.
Congratulations to @Alibaba_Qwen on the release!
See below for our analysis and example outputs of Qwen-Image-2.1 in the Artificial Analysis Image Arena 🧵