Alibaba’s Qwen team released Qwen-Image-2.1, a new image generation and editing model with 7 billion parameters. The model supports native transparent image creation and can work with up to 10 reference images in a single workflow. It was announced on September 20.
The new model combines image generation and editing in one system. Users can create images from text and modify existing visuals without switching between separate models. Qwen also added support for native RGBA images and 2K output.
One of the key additions is support for up to 10 reference images. Users can provide several images and ask the model to combine their visual information in a single editing task. This can support workflows such as group portraits, product visuals and room designs. The model also allows local edits through masks, painted marks or selected areas. This lets users target a specific part of an image while keeping other areas unchanged.
Qwen said that the model uses a 32-layer single-stream diffusion transformer for visual generation. It also uses a Qwen3-VL 8B text encoder and a 64-channel RGBA autoencoder. The company reported a score of 60.28 on its Qwen-Image-Bench evaluation. That score is higher than the listed field average of 59.82. However, six closed models scored higher in the same comparison. Qwen’s figures come from its own benchmark and had not been independently reproduced at launch.
Qwen-Image-2.1 is available for download through Hugging Face, ModelScope and GitHub. The model also consists of day-one support from tools including ComfyUI, Diffusers, vLLM-Omni and SGLang. However, the release comes with a research licence. The public licence allows non-commercial use, while commercial users need a separate arrangement with Qwen.
This marks a change from earlier Qwen-Image releases that used Apache 2.0 licensing. The model is designed to run on capable consumer hardware, including GPUs such as the RTX 3090, according to Qwen’s documentation. Exact performance depends on the configuration and workflow used.
Also Read: Alibaba to Raise USD 10.2 Billion to Accelerate AI Expansion