Alibaba’s Qwen team has released downloadable Turbo weights for generating and editing images. The checkpoint uses an eight-step denoising schedule and works with the Diffusers library. Free use covers research and evaluation, while commercial deployment requires a separate agreement. Hosted Pro and Turbo interfaces have also opened through Alibaba Cloud’s Model Studio.
Alibaba’s Qwen AI team released the weights of Qwen-Image-2.1-Turbo on 9 October. The model creates images from text and edits existing images from natural-language instructions. Developers can download it through Hugging Face and ModelScope, two platforms distributing machine-learning models. The release follows the original Qwen-Image-2.1 model published on 20 September.[1], [2]
Image creation uses 8 denoising steps, progressively removing noise to form the picture. Diffusers, a library for running image models, loads the recommended schedule with the checkpoint. Changing the step count alone does not replace that saved schedule; a different schedule must be supplied explicitly. Shared text and reference-image context can be cached between steps to reuse computation.[1]
The Qwen Research License permits free research and evaluation. Commercial use requires a separate agreement, a condition that also applies to these downloadable weights. The repository’s examples include transparent-image generation, editing transparent layers and extracting a subject from a photograph. Qwen has also opened hosted Pro and Turbo programming interfaces through Alibaba Cloud Model Studio, giving applications a hosted route alongside local model workflows.[1], [2]
