Qwen(@Alibaba_Qwen)

Thanks @sgl_project for the day-0 support! 🙌 SGLang-Diffusion now serves Qwen-Image-2.1: text-to-im...

6.5内容质量
Thanks @sgl_project for the day-0 support! 🙌 SGLang-Diffusion now serves Qwen-Image-2.1: text-to-im...

TL;DR · AI 摘要

通义千问宣布SGLang-Diffusion支持Qwen-Image-2.1模型，提供文本生成、图像编辑等功能，但技术细节披露有限。

核心要点

  • Qwen-Image-2.1支持透明RGBA输出和多图像编辑功能
  • RTX 4090 24GB单卡生成1024×1024图像需18.7秒
  • Qwen3.8-Flash模型参数量达125B但每token仅激活6B参数

结构提纲

按章节快速跳转。

  1. §技术发布声明

    阿里巴巴宣布SGLang-Diffusion平台新增Qwen-Image-2.1模型支持

  2. ·核心功能升级

    新增文本到图像生成、多图像编辑及透明通道输出能力

  3. ›性能基准数据

    在RTX 4090 24GB显卡上实现18.7秒/张的生成速度

  4. ·用户反馈

    社区对许可证政策和模型效率提出讨论与建议

  5. ›模型优化进展

    Qwen3.8-Flash实现125B参数量与6B激活参数的平衡

思维导图

用一张图看清主题之间的关系。

查看大纲文本（无障碍 / 无 JS 友好）
  • Qwen-Image-2.1技术发布
    • 核心功能
      • 文本生成
      • 多图像编辑
      • 透明通道输出
    • 性能指标
      • RTX 4090 18.7s
      • RTX 6000 8.0s
    • 用户反馈
      • 许可证政策讨论
      • 模型效率评价

金句 / Highlights

值得收藏与分享的关键句。

#AI绘画#模型优化#图像生成#显卡性能
打开原文

[](https://x.com/)

Qwen on X: "Thanks @sgl_project for the day-0 support! 🙌 SGLang-Diffusion now serves Qwen-Image-2.1: text-to-image generation, multi-image editing, and transparent RGBA output. Try it out! 🎨"

@sgl_project for the day-0 support! 🙌 SGLang-Diffusion now serves Qwen-Image-2.1: text-to-image generation, multi-image editing, and transparent RGBA output. Try it out! 🎨 ![Image 2: @sgl_project](https://x.com/sgl_project) Day-0 support for

@Alibaba_Qwen ’s Qwen-Image 2.1 is here in SGLang-Diffusion! 🖥️ Native precision on a single RTX 4090 24GB with CPU offload - 1024×1024 generation in 18.7s and image editing in 21.7s with 22.7 GiB peak GPU memory during requests. - On an RTX PRO 6000 96GB: 8.0s

  • ![Image 3: @bayeschat](https://x.com/bayeschat) Single-GPU support with CPU offload makes this much easier to try locally. Native 2K generation plus editing in one checkpoint is a very practical combination.
  • ![Image 4: @karimabaz](https://x.com/karimabaz) Will you be updating your licensing policy? It's really restrictive.
  • ![Image 5: @rb_walter_ai](https://x.com/rb_walter_ai) Qwen3.8-Flash looks like a massive step forward for open-weight models! Balancing 125B parameters with just 6B activated per token while keeping training costs so low is incredible efficiency. ​(For paid partnerships or business collaborations, feel free to send a DM!)