Qwen-Image-2.1-Turbo: 8-step generation and editing, non-commercial licence

Qwen released Qwen-Image-2.1-Turbo on October 9, 2026, according to the News section of the QwenLM/Qwen-Image-2.1 GitHub repository. It is an accelerated checkpoint of Qwen-Image-2.1, released September 20, 2026, and uses the same 7B visual generation architecture. The 7B figure covers that component only. The README lists a Qwen3-VL 8B text encoder for Qwen-Image-2.1, and the Turbo repository also ships a Qwen3-VL text encoder. Turbo does text-to-image generation and image editing in 8 denoising steps, with CFG=1 by default. The checkpoint carries its own sampling schedule. Setting `num_inference_steps` alone does not override it, and the model card says other schedules have not been evaluated. Also on October 9, Qwen said hosted Qwen-Image-2.1 Pro and Turbo APIs are live on Alibaba Cloud Model Studio. Qwen says Qwen-Image-2.1 natively generates transparent (RGBA) images, and the Turbo model card on Hugging Face shows a transparent image example. The Hugging Face page lists the weights as BF16 safetensors. They load through `QwenImage21Pipeline` in Diffusers installed from source. The model card says the checkpoint needs Diffusers support for pipeline-configured sampling sigmas, added in PR #14950. For Apple Silicon, the README points to LightX2V-Mac-App, a native macOS app in the ModelTC GitHub organisation, built on the LightX2V framework. Neither the README nor the model card gives a VRAM figure. For the Diffusers pipeline, the README's memory advice is model offloading on GPUs with limited memory. The README lists day-0 ComfyUI support for the base Qwen-Image-2.1, but does not mention ComfyUI for Turbo. The licence is the Qwen Research License Agreement. It grants rights for non-commercial purposes only, and defines non-commercial as research or evaluation. Commercial use needs a separate commercial licence from the licensor, Hangzhou Tongyi Laboratory Technology Co., Ltd.