Alibaba's open-weight Qwen-Image-2.1 claims to beat closed models in image generation with just 7 billion parameters
decoder ·
Alibaba’s Qwen AI team has released Qwen-Image-2.1, an open-weight model for image generation and editing. The model’s visual generation component has 7 billion parameters, and Qwen claims it outperforms most closed models on the team’s own benchmark, although independent evaluations are not yet available. It can run on capable consumer GPUs such as the Nvidia 3090. The model supports native generation and editing of transparent RGBA images, up to ten reference images, and localized edits guided by circles, masks, or painted marks. Qwen-Image-2.1 is available through Hugging Face, GitHub, and Model Scope, but its research license excludes commercial use unless businesses obtain a separate license from Qwen.
decoder