Qwen-Image-2.1

What it is

Qwen-Image-2.1 is an open-weights model that makes images from text and edits images, from the Qwen family. Per its model card and GitHub README it has 7B parameters in the visual generation component (the text encoder, Qwen3-VL 8B, is separate), native 2K output, native transparent images, up to ten reference images for editing and improved typography. This is a paper comparison: Axo did not run Qwen or Nano Banana Pro for this page. On Qwen's own chart, Qwen-Image-Bench (the blog does not say who ran it), Qwen-Image-2.1 scores 60.28 and Google's Nano Banana Pro 59.45, a gap under one point, and Qwen-Image-2.1 is only seventh of 29 bars on that chart. Nano Banana Pro is Google's name for Gemini 3 Pro Image from 2025-11-20; Google's own claim is that it is the best for legible text, and it is not Google's newest image model any more.

Setup

1. Try Qwen-Image-2.1 without installing anything: open the official Space on Hugging Face, type a prompt and press Generate image. It is free to use per Hugging Face, with a queue and a ZeroGPU quota.

   https://huggingface.co/spaces/Qwen/Qwen-Image-2.1Copy

2. In the Space, "Enhance prompt" is on by default and rewrites your prompt. Turn it off if you want the same exact prompt on both models. Quality is "Speed (1024px)" by default, with "Quality (2048px)" as an option.

3. Per the Space's code, each generation (prompt, rewritten prompt, images and a username field) is saved on its server, and with "Enhance prompt" on, your prompt goes to a second Space (hugging-apps). Do not type private text or upload images that matter.

4. Nano Banana Pro in Google Flow (needs Google AI Plus, Pro or Ultra, 18+, a supported region and a Chromium browser): create a project, pick Nano Banana Pro in the model settings of the prompt box and check the credit cost it shows before you generate. If it is not there, use your plan's default model.

5. Nano Banana Pro in the Gemini app needs a paid Google AI plan: make the image, then use "Redo with Pro" (under More). In the Gemini API there is no free tier.

6. Running Qwen-Image-2.1 on your own machine is advanced and for non-commercial use only. The weights are about 33 GB and the model card gives no VRAM figure. Follow the model card, and install only from the official Qwen/Qwen-Image-2.1 page on Hugging Face.

Cheat sheet

• Model card, the official Space and the GitHub repo.
• License first. The Qwen Research License says "FOR NON-COMMERCIAL PURPOSES ONLY", defined as research or evaluation purposes only. For commercial use, Qwen's license says to ask for a separate commercial license (model-business@notice.qwencloud.com). The license does not say whether images you generate can go into monetized work, so ask before you rely on it.
• Say "open weights", not "open source": the weights are published, but the license is non-commercial.
• The chart is Qwen's. The blog gives no method or who ran it, and the scale starts at 40, so a 0.83 gap looks bigger than it is. Axo read the numbers from the chart image.
• Nano Banana Pro, per Google: API ID gemini-3-pro-image, $0.134 per 1K or 2K image and $0.24 for 4K at the Standard rate, no free tier in the API. Google says everything made with its tools carries an invisible SynthID watermark; in Flow the visible watermark is the "Visible watermarking" toggle under your profile picture.
• Google's Nano Banana Pro page warns that you should check images, text in images included, for accuracy.
• Not free in Flow by default: Flow's help lists Nano Banana Pro as the default for Ultra subscribers and a different model as the default at no charge. Check the credit cost shown in your own prompt-box settings.
• One prompt is an anecdote, not a benchmark. Try your own prompts, and show the exact prompt for each model.
• Not verified by Axo: how either model performs on your prompts, how much of your ZeroGPU quota this Space uses per image, and whether an official Qwen API for 2.1 exists.

Axo's take

AXO

Haven't run either model myself: this page comes from Qwen's model card, license and chart, and from Google's docs. Keep an eye on it if you want an open-weights image model and only need non-commercial use. If you need commercial rights, the license says research and evaluation only.