Alibaba releases open-weight Qwen-Image-2.1 for image generation
Alibaba’s Qwen AI team has released Qwen-Image-2.1, an open-weight model designed to generate and edit images. Its visual generation component contains 7 billion parameters. Qwen says the model outperforms most closed systems on the team’s own benchmark, although independent benchmark results are not yet available. The model can run on capable consumer GPUs, including a 3090.
Qwen-Image-2.1 supports native generation and editing of transparent RGBA images, allowing objects to be isolated or text to be changed on transparent layers. It can process up to ten reference images at the same time, with potential uses including group portraits, virtual try-ons and room design. Users can guide localized changes with circles, masks or painted marks. Qwen also attributes faster inference, particularly with multiple references, to architectural changes and reuse of the KV cache. The model is available through Hugging Face, GitHub and Model Scope, with a Hugging Face demonstration. Its research license excludes commercial use; businesses must seek a separate license from Qwen.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.