
9月20日,阿里通义千问正式推出 Qwen-Image-2.1 图像模型。该模型将文字生图与图像编辑统一到同一架构中,视觉生成模块拥有 7B 参数,原生支持 2K 分辨率与 RGBA 透明图。
新模型可接受最多 10 张参考图,并提供圈选、涂画、Mask 等局部修改方式,创作者可在同一模型内完成从生成到精细化编辑的闭环,显著降低多工具切换成本。
Qwen-Image-2.1 首日即支持 ComfyUI 和 Diffusers 工作流,延续千问系列的开源策略,进一步降低了高品质图像生成与细节编辑的门槛。
在 AIGC 日报当天简讯中,千问此次升级与阶跃星辰 Step 5 Preview、硅基流动融资一同被列为 9 月 20 日 AI 领域的关键进展,彰显国内大模型公司在模型能力与开发生态上的密集迭代。
统一的生图与编辑架构契合创作者对“一处完成、所见即所得”的工作流需求,也为电商、设计、短视频等场景的批量素材生产提供更轻量的本地化方案。
本文来源:http://placeholder.local/article_qwen
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
On September 20, Alibaba's Tongyi Qianwen officially launched the Qwen-Image-2.1 image model. The model unifies text-to-image and image editing into a single architecture, with a 7B-parameter vision module that natively supports 2K resolution and RGBA transparency.
The new model accepts up to 10 reference images and provides localized modification options such as region selection, painting, and masking. Creators can complete the full cycle from generation to fine-tuning within a single model, significantly reducing the overhead of switching between tools.
Qwen-Image-2.1 is compatible with ComfyUI and Diffusers workflows on its first day, continuing the open-source strategy of the Qianwen series. This further lowers the barrier to high-quality image generation and detailed editing.
In the AIGC Daily news brief, this upgrade was highlighted alongside Step 5 Preview and Silicone Flow financing as a key development in the AI field on September 20, demonstrating rapid iteration by domestic model companies.
The unified architecture for generation and editing meets creators' demand for an "all-in-one, what-you-see-is-what-you-get" workflow, offering a lighter, localized solution for mass material production in e-commerce, design, and short video contexts.