Back to feed
News Story
量子位
1 sources

Qwen opens weights for 7B image generation model Qwen-Image-2.1, runs on RTX 3090, supports 2K output and editing

Alibaba's Qwen has released the model weights for Qwen-Image-2.1, a 7B image generation model that supports text-to-image, image editing, transparent image generation, and native 2K output, and can run on consumer GPUs like the RTX 3090. After release, it topped Hugging Face's text-to-image trending list, and official comparisons show it outperforming models like Nano Banana 2.0 and GPT Image 1.5. However, it uses the Qwen Research License, meaning the open weights are limited to non-commercial research and evaluation by default, with commercial use requiring separate authorization.

SynthePulse Insight · AI deep readingMembers

7B Image Model Open Weights: Where Are the "Open" Boundaries of Qwen-Image-2.1?

Version 1 · 1 source

Qwen has opened the weights of its 7B image generation model Qwen-Image-2.1, which can run on an RTX 3090, and it topped the Hugging Face text-to-image trending chart. But this "openness" uses the Qwen Research License, which by default permits only non-commercial research and evaluation—the gap between capability and licensing is what truly makes this release worth reading about.

  • Qwen-Image-2.1 is a 7B image generation model with open weights that can run on an RTX 3090.
  • The same set of weights covers text-to-image, image editing, transparent image generation, and native 2K output, with up to 10 reference images.
  • A user tested it on dual RTX 4090s with 48GB VRAM: about 21 seconds for a standard high-definition image, and over two minutes for a 2752×1536 2K image, without running out of VRAM.
Open section navigationOne Release, Two Keywords: 7B and Open Weights

One Release, Two Keywords: 7B and Open Weights

This time Qwen has opened the weights of Qwen-Image-2.1, a 7B-scale image generation model that, per the official statement, can run on an RTX 3090. After release, it quickly topped the Hugging Face text-to-image trending chart. For readers interested in local deployment, the 7B scale plus consumer-grade GPU compatibility is the most direct signal from this news.

But "open" does not mean "use however you want." According to QbitAI's report, Qwen officially uses the Qwen Research License rather than a permissive open-source license like Apache 2.0; under the agreement, the public weights by default can only be used for non-commercial research and evaluation, and enterprises wanting to integrate them into commercial products need to separately apply to Qwen for authorization. Capability openness and commercial licensing have been split into two separate matters—this is the premise for understanding this release.

Free for now

Read the full analysis

4 more sections of analysis, plus the full takeaway

Loading

Credibility boundary

All information in this article comes from a single QbitAI report on the release of Qwen-Image-2.1, and is secondhand paraphrase. Among it, the model scale, license type, Hugging Face trending chart ranking, and official comparison image conclusions are official or platform information paraphrased by the report; the generation times of 21 seconds and over two minutes come from a single user's self-test in a dual RTX 4090, 48GB VRAM environment, with a limited sample; "surpasses Nano Banana 2.0 and GPT Image 1.5" is the official comparison image's claim and has not been independently verified.

Sources for this version

  1. 千问7B生图模型开放权重！3090就能跑，2K出图修图抠图全包量子位

Primary report

量子位

Primary source