Alibaba releases Qwen-Image-2.1 for creating and editing transparent images
The Qwen-Image-2.1 model from Alibaba natively generates and edits transparent images, creates images in 2K and processes up to ten reference images. Its weights are available, and it also runs on an NVIDIA 3090 GPU. Commercial use requires a separate license.
Alibaba released the open-weight model Qwen-Image-2.1 for generating and editing images. Its visual generative component has approximately seven billion parameters and creates images natively in 2K. It supports transparency in RGBA format, allowing users to create isolated objects or change text on transparent layers.
When editing, the model works with up to ten reference images simultaneously, for example for group portraits, virtual clothing try-ons or room designs. Local edits can be guided using circles, masks or drawn marks. According to Alibaba, the model outperforms most closed models in its own comparative test; independent verification of this claim is still lacking.
The model is available on Hugging Face, GitHub and Model Scope, with a demo also available on Hugging Face. It can run on powerful consumer GPUs, such as NVIDIA 3090. The research license prohibits commercial use; business users must request a separate license from the Qwen team.
Why it matters
Graphic designers can create objects and edit text directly on transparent layers that can be used in other compositions. The available weights allow local operation on the specified GPU. However, obtaining a separate license is an essential condition for commercial graphics production.
Release card
Qwen-Image-2.1
Alibaba
- Specifications
- The visual generative component has approximately 7 billion parameters.
- Inputs
- The model generates and edits images, including transparent images in RGBA format. When editing, it processes up to ten reference images simultaneously and creates images natively in 2K.
- Availability
- The model is available on Hugging Face, GitHub and Model Scope. There is also a demo on Hugging Face. It can run on powerful consumer GPUs, such as NVIDIA 3090.
- Allows users to create isolated objects and change text on transparent layers.
- Allows users to use multiple reference images for group portraits, virtual clothing try-ons or room designs.
- Allows users to guide local edits using circles, masks or drawn marks.
- The research license does not allow commercial use; that requires a separate license from the Qwen team.
According to the Qwen team, the model outperforms most closed image generation models in its own comparative test. Independent measurements are still lacking.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
What was added since the original report
Verified updates
-
The model generates images natively at 2K resolution.
Two audiences, two different impacts
What this means
For individuals
When working with graphics, you can generate transparent objects and edit text on transparent layers; up to ten reference images can be combined when editing.
For a business
Deployment in commercial graphics production requires a separate license. The published research license does not allow a company to use the model commercially.
Risks and complianceCheck the original
Event sources
confirmed by 2 independent sources · 2 publishers, 2 independent. We count feeds from the same owner only once.