The fastest way to get this model running locally is via Docker.
Please follow the instructions listed below to get started.
Following this guide to the end unlocks everything you ever wanted to get out of this environment.
Z-Image-Turbo is a next‑generation AI image generation model designed for **ultra‑fast inference** while preserving **high visual fidelity**. It leverages a novel **spatially‑adaptive denoising** architecture that reduces computational overhead by up to 70% compared to previous models. The model supports native resolutions up to **4K** and can generate a full‑frame image in under **200 ms** on a single GPU. Integration with popular pipelines is streamlined through a unified API that accepts text prompts, style references, and control nets. A comparison table below highlights its performance against leading competitors, showcasing superior speed‑quality trade‑offs.
| Metric | Z-Image-Turbo | Competitors |
|---|---|---|
| Inference Time | < 200 ms | 300‑500 ms |
| Max Resolution | 4K | 2K‑3K |
| Parameters | 1.5 B | 2‑3 B |
| GPU Memory | 8 GB | 12‑16 GB |