Alibaba's Qwen team has released Qwen-Image-2.1, a 7-billion-parameter diffusion transformer that unifies text-to-image generation, multi-reference editing, and native RGBA transparency in a single checkpoint. The weights are public, making the model open-weight.
A notable feature is a prefix KV cache that speeds up editing workflows when working with up to 10 reference images. This design allows the model to handle multiple inputs in one pass, avoiding the need for separate fine-tuned models for each task.
Because the source is a single announcement, there are no conflicting details to compare. The release positions Qwen-Image-2.1 as a multi-purpose image tool, though the source does not include benchmark numbers or comparisons with other models.