Qwen Image 2.1: what it does, the license catch and where to use it
By GetStepSnap · Checked against Qwen's and Alibaba Cloud's own pages on October 19, 2026
- Is it free?
- The weights are, for research and evaluation. Paid work needs a commercial license from Alibaba.
- Where can I use it?
- Qwen's demo on Hugging Face, Alibaba Cloud's API, or your own computer.
- What does the API cost?
- $0.04 an image for Pro and $0.016 for Turbo, on Alibaba Cloud's international site.
- Can it run on my PC?
- Yes, with Diffusers or ComfyUI. Qwen also points to a Mac app and Intel Arc support.
Qwen Image 2.1 is an open-weights image model from Alibaba's Qwen team, out since September 20, 2026. It makes pictures from a sentence and edits the ones you give it, in one model, and the part that draws is only 7 billion parameters. Qwen points to ways of running it on a Mac or on Intel graphics, not just on a big Nvidia card.
The catch is the license. The weights are free to download, but they come under the Qwen Research License, which covers research and evaluation only. Client work, a product, an app that sells images: all of that needs a separate commercial license from Alibaba, or Alibaba Cloud's paid API. Qwen's model card calls the release open source, which is generous wording for a research-only license.
What's new in Qwen Image 2.1
- Transparent images, built in. Ask for a transparent sticker or logo and it returns a picture with a real transparent background. It also cuts a subject out of a photo and edits transparent layers, text included.
- Up to 10 reference images. Qwen's own examples put six portraits into one group photo, and a model, clothes, shoes, a bag and a hat into one outfit.
- Point at what to change. Circle a spot, paint over it or send a separate mask, and the edit stays inside it. Faces and product labels are meant to survive the edit.
- Better text and portraits. Qwen says it improved typography, portrait lighting and fine detail over earlier Qwen Image models.
- A Turbo version. Out October 9, 2026, it makes or edits an image in 8 steps, where Qwen's examples for the full model use 40. Same license.
The license catch
The Qwen Research License grants the weights for non-commercial purposes only, and defines non-commercial as research or evaluation. Using them for any commercial purpose needs a separate license, which you ask Alibaba for by email. If you train and release a model with it, you also have to say "Built with Qwen" or "Improved using Qwen".
So the free version is for testing it, learning it, or comparing it with other models. For paid work there are two routes: Alibaba's commercial license, or the hosted API on Alibaba Cloud Model Studio, a paid service that runs under Alibaba Cloud's own terms rather than the research license.
Where to use it
| Where | Cost | Good for |
|---|---|---|
| Qwen's demo on Hugging Face | Free | Trying a few prompts |
| Qwen Chat (chat.qwen.ai) | Free with an account | Trying it, though its image tool didn't say it runs 2.1 when we checked |
| Your own computer | Free download; your graphics card does the work | Research, testing, learning ComfyUI |
| Alibaba Cloud Model Studio API | $0.04 an image (Pro), $0.016 (Turbo) | Apps and paid work, under Alibaba Cloud's own terms |
API prices are Alibaba Cloud's list prices for its international (Singapore) site on October 19, 2026, before promotions. The Beijing region lists a little less.
Running it on your own computer
The weights are on Hugging Face and ModelScope. Diffusers, the Python library, and ComfyUI, the node-based app many people use for local image models, both supported it on day one, and Comfy-Org publishes example workflows for text-to-image and editing. Turbo is the version to start with if your card is slow: 8 steps, where the full model's examples use 40.
Qwen doesn't list a minimum graphics card. Its Diffusers code has a memory-saving mode that moves parts of the model to the CPU, which fits a smaller card at the cost of speed, and community builds shrink the files further. On a Mac with Apple Silicon, Qwen points to the LightX2V app; Intel supports it on Arc graphics through OpenVINO. Whatever you run locally is still under the research license.
Need images for paid work today?
Qwen Image 2.1 isn't on GetStepSnap. What is: Nano Banana 2.1 and GPT Image 2.5 for pictures from a sentence or edits of your own photo, and FLUX 3 for pictures from text. No subscription: packs cost $4.90 for 20 credits or $9.90 for 70, paid once. For another new editing model, see what Krea 3 costs and where it runs.
Try Nano Banana 2.1Questions
Is Qwen Image 2.1 free?
The weights are free to download from Hugging Face and ModelScope under the Qwen Research License, which allows research and evaluation only. Qwen's demo on Hugging Face costs nothing to use. For paid work you need a commercial license from Alibaba, or its hosted API, which charges per image.
Can I use Qwen Image 2.1 commercially?
Not with the free weights alone. The license says any commercial use needs a separate commercial license from Alibaba, and gives an email address to ask for one. Alibaba Cloud's hosted API is a paid service that runs under Alibaba Cloud's own terms.
When did Qwen Image 2.1 come out?
September 20, 2026, with the weights and day-one support in Diffusers and ComfyUI. A faster Turbo version and the hosted Pro and Turbo APIs followed on October 9, 2026.
How much does the Qwen Image 2.1 API cost?
On Alibaba Cloud Model Studio's international site (Singapore), $0.04 an image for qwen-image-2.1-pro and $0.016 for qwen-image-2.1-turbo, at list price on October 19, 2026. Prices in other regions differ a little.
What graphics card do I need to run Qwen Image 2.1?
Qwen doesn't publish a minimum. The image part has 7 billion parameters, and its Diffusers pipeline can move parts of the model to the CPU to fit a smaller card, more slowly. Community builds shrink it further. Qwen also points to the LightX2V app for Apple Silicon Macs and to Intel's OpenVINO for Intel Arc graphics.
Can Qwen Image 2.1 make transparent images?
Yes, and it's the headline feature. Ask for a transparent image in the prompt and it returns one with an alpha channel. It can also cut a subject out of a photo and edit a transparent layer without losing the transparency.
Can I use Qwen Image 2.1 on GetStepSnap?
No. The free weights are for research only, so offering it in a paid tool would take Alibaba's commercial license or its paid API, and we haven't added it. For pictures from a sentence or edits of your own photo, our Nano Banana 2.1, GPT Image 2.5 and FLUX 3 tools do the same kind of job.
Sources, checked October 19, 2026: Qwen-Image-2.1 on Hugging Face, the Qwen Research License, Qwen-Image-2.1 on GitHub, Qwen's blog post, Alibaba Cloud: qwen-image-2.1-pro, Alibaba Cloud: qwen-image-2.1-turbo.
