ahadadada/image2image-with-z-image-turbo

Image-to-Image with Z-Image Turbo

The first Hugging Face Space running Tongyi-MAI/Z-Image-Turbo with a full Image-to-Image workflow (strength control, seed, batch, negative prompt). Existing public Z-Image Spaces are text-to-image only.

Features

• —Image-to-Image — upload any image, pick denoising strength (0–1), describe the transformation. Aspect ratio is preserved (contain-fit, no squash).
• —Text-to-Image — 8-step Turbo generation, CFG control (FlowMatch Euler scheduler).
• —Seed control + random seed, batch size up to 4, negative prompt.
• —Live per-step progress bar.
• —Model: Tongyi-MAI/Z-Image-Turbo (Apache-2.0), loaded automatically at startup (~11 GB download, cached by HF).

Tips

• —For subtle edits use strength 0.3–0.5; for full restyling 0.7–1.0.
• —Turbo is a distilled model: keep CFG = 0.0 (higher CFG degrades quality).
• —8 steps is the native setting; more steps give diminishing returns.