Qwen-Image-2.1-Turbo

Qwen-Image-2.1-Turbo is an accelerated checkpoint of the Qwen-Image-2.1 model developed by Alibaba, designed for rapid text-to-image generation and image editing. Released on October 9, 2026, the model cuts down the required sampling steps significantly, performing high-quality generation and editing tasks in just eight denoising steps compared to the base model's standard trajectory.

Built around the same 7B parameter visual generation component featuring 32 Single-Stream DiT layers, the Turbo checkpoint preserves the architecture and core capabilities of its predecessor. It retains unified support for text-to-image generation, native transparency (RGBA) creation, multi-reference composition using up to 10 reference images, and local editing through intuitive user prompts or masks.

Efficiency improvements are achieved via the streamlined 8-step sampling schedule bundled directly into the checkpoint, alongside default configurations such as CFG=1 and prefix KV cache reuse for text and reference context. The model maintains strong fidelity for detailed textures, accurate typography, and complex visual scenes while dramatically lowering inference times for iterative workflows.