Alibaba's open-weight Qwen-Image-2.1 claims to beat closed models in image generation with just 7 billion parameters

AI 导读

阿里推出开源图像生成模型Qwen-Image-2.1，仅70亿参数即超越多数闭源模型，支持透明图像编辑与多图合成，但商业使用需单独授权。

Alibaba's Qwen-Image-2.1, an open-source image generator with 7B parameters, claims to outperform most closed models, though commercial use requires separate licensing.

重点速览

• Qwen-Image-2.1以70亿参数超越多数闭源模型 Qwen-Image-2.1 outperforms most closed models with 7B parameters
• 支持透明图像编辑及多张参考图合成 Supports transparent image editing and multi-image synthesis
• 需通过Qwen申请商业使用许可 Requires separate licensing for commercial use
• 可运行于消费级GPU如RTX 3090 Runs on consumer GPUs like RTX 3090
• 提供Hugging Face等多平台开源版本 Available on Hugging Face and other platforms

一句话：开源大模型在图像生成领域展现竞争力，但商业化仍需合规授权。 / Open-source models demonstrate competitiveness in image generation, yet commercialization requires licensing.

推荐理由 展示开源大模型在性能与效率上的突破，为AI研究提供新范式。