Qwen-Image-2.1-Turbo

Text-to-image generation and multi-image editing with Qwen/Qwen-Image-2.1-Turbo, the official 8-step distilled checkpoint of Qwen-Image-2.1, running on ZeroGPU through QwenImage21Pipeline in diffusers.

• Text to image: leave the input gallery empty and describe what you want.
• Image editing: upload up to 10 images and refer to them as <image1> … <image10>.
• Transparency: the model natively decodes RGBA. Ask for it in the prompt.

Sampling follows the model card: the checkpoint's saved 8-step sigma schedule, CFG 1 and prefix KV caching.

Speed

The transformer blocks and VAE decoder run as ahead-of-time compiled kernels (AOTInductor) from hugging-apps/qwen-image-2-1-aoti. Turbo has the same architecture as Qwen-Image-2.1, and weights are runtime inputs to the compiled graph, so the same artifacts apply. Set QWEN21_AOTI=0 to run fully eager.

Prompt enhancement

The optional Enhance prompt checkbox calls hugging-apps/qwen-image-2-1-prompt-enhancer, which serves the official Qwen-Image-2.1-PE-T2I / PE-I2I rewriters.

Safety

Editing requests (any submission with input images) are screened with hfmlsoc/ncii-guard-v02 before anything is generated.

Example assets

examples/turbo-boat-input.png and the two "model card" prompts come from the Qwen-Image-2.1-Turbo model card. The other images under examples/ are official demo assets from the Qwen/Qwen-Image-2.1 Space. All are redistributed under the Qwen Research License for non-commercial demonstration.

License

Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) 2026 Hangzhou Tongyi Laboratory Technology Co., Ltd. All Rights Reserved.

Research and evaluation use only. See the license.