Alibaba's Qwen team released Qwen-Image-2.1-Turbo on October 9, 2026, as an accelerated version of Qwen-Image-2.1, which was open-sourced on September 20. According to the official repository, Turbo uses the same 7B parameter visual generation architecture with 32 single-stream DiT layers and supports both text-to-image generation and image editing, but finishes in only 8 denoising steps.
The Turbo weights are published on Hugging Face, and the checkpoint ships with its recommended sampling schedule, which Diffusers loads automatically through QwenImage21Pipeline with a default CFG of 1. Qwen notes that changing the step count alone does not override that saved schedule. On the same day, Qwen-Image-2.1 Pro and Turbo APIs went live on Alibaba Cloud Model Studio for teams that would rather not host the model themselves.
The Qwen-Image-2.1 family can generate regular or transparent RGBA images from text, edit transparent layers, extract subjects from photos, and edit with up to 10 reference images. Users can mark local edits with circles, painted annotations or separate masks while preserving the identity of people and products. Qwen also says this generation improves typography, portrait lighting and fine detail.
The Turbo model card lists the Qwen Research License, so anyone planning commercial work should review the license terms first. For web and design work, transparent assets, multi-image composition and local retouching map directly onto hero images and product pages.