⚡ Qwen-Image-2.1-Turbo
8-step text-to-image and reference-image editing with the official distilled checkpoint of Qwen-Image-2.1.
Status: waiting for GPU hardware.
The Gradio app for this demo is complete and sits in this repository (
app.py), written for ZeroGPU.
It goes live once a GPU community grant is attached; until then this page is a placeholder.
The same model can be tried right now in the
hugging-apps/qwen-qwen-image-2-1-turbo Space.
What the demo does
- Text to image at 1K, 1.5K or 2K in the model card's seven aspect ratios.
- Editing and composition with up to four reference images, addressed as
<image1>…<image4>in the prompt. - Transparent output: the model decodes RGBA, so a transparent background can be requested in the prompt.
- Sampling exactly as the model card prescribes: the checkpoint's saved 8-step sigma schedule, CFG 1, prefix KV cache.
- Editing prompts are screened on CPU by hfmlsoc/ncii-guard-v02 before any GPU time is reserved.
- Exposed as an MCP server through the
generateendpoint.