Quick answer
Qwen released Qwen-Image-2.1 on September 20 with downloadable weights for image generation and editing, including transparent images. The important qualification is its Research License Agreement: the published grant is for research or evaluation, and commercial use of the model materials requires a separate license. This is a source-based release explainer, not a hands-on quality test.
Transparency is an output feature, not just a white background
Qwen's model card describes native RGBA generation, editing of transparent layers, and subject extraction from photographs. RGBA adds an alpha channel that controls opacity, so a cutout can sit over another background. A picture of a checkerboard or a white backdrop is not equivalent to transparency in the exported file.
- The same model handles text-to-image creation and image-conditioned editing.
- Qwen documents up to 10 reference images, with local edits indicated by annotations or masks.
- The repository lists a 2048 × 2048 square output among its recommended sizes.
Sources: Qwen Team on Hugging Face, Qwen Team
Downloadable weights do not grant commercial permission
The license defines non-commercial use as research or evaluation only. It requires a separate commercial license to use the model materials commercially. Do not treat a public download as permission to deploy the model in a paid product or client-production workflow.
Sources: Qwen Team
Access exists, but the 7B label is not a device requirement
The release repository links downloadable weights on Hugging Face and ModelScope, a demo, and Diffusers and ComfyUI workflows. Its architecture notes put 7B parameters in the visual-generation component and list a separate Qwen3-VL 8B text encoder. The 7B label alone therefore does not establish the full memory requirement or prove that the model will run on a phone.
Sources: Qwen Team
An editorial exercise for checking a transparent cutout
For a research or evaluation session in a tool that explicitly supports Qwen-Image-2.1, use an original subject and inspect the exported file. This proposed exercise is not a live model output or a test performed in Chat AI.
- Export the original result as a PNG with transparency, rather than taking a screenshot of the preview.
- Place it over both a pale background and a dark background in an image editor. Check whether the surrounding area reveals the background instead of showing a baked-in white or checkerboard rectangle.
- Inspect the folded edges for unwanted halos, missing pieces, and opaque patches. Then request only a color change from orange to blue and compare the shape and edges again.
Example prompt — select and copy
Create an RGBA image of a small orange paper boat, viewed from slightly above. Keep the whole boat inside the frame. Leave the surrounding area transparent, with no scenery, lettering, or cast shadow. Preserve the folded paper edges.
Frequently asked questions
What readers usually ask
Is Qwen-Image-2.1 unrestricted for commercial use?
No. The published Research License Agreement limits its grant to research or evaluation and requires a separate license for commercial use of the model materials. Hosted-service terms and output rights should be checked separately.
Does a PNG file automatically have a transparent background?
No. PNG can preserve transparency, but an exported PNG can also be fully opaque. Inspect it over contrasting backgrounds and check that the surrounding pixels are actually transparent.
Does this release confirm Qwen-Image-2.1 is in Chat AI?
No. This article covers Qwen's release. The exact model is not verified in Chat AI's published catalog, and the exercise above is not a Chat AI test.
Evidence