Z Image Turbo

What is Z Image Turbo?

Z Image Turbo is a hosted workspace around Tongyi-MAI’s open Z-Image-Turbo 6B S3-DiT text-to-image model, distilled to an 8-step pipeline. On the managed GPU cluster most scenes finish in about 2–4 seconds; the same bilingual English/Chinese checkpoint can also run via REST API or ComfyUI with Apache-licensed weights. The site adds credits, stored prompts, team folders, and a growing multi-model workspace for image and related video tasks. Free starter credits cover trials; paid plans increase volume and priority. Recommended sampler settings keep CFG near 1.0, Shift between 3 and 6, and steps around 8–10.

How to use Z Image Turbo?

1. Open zimageturbo.ai and enter an English or Chinese prompt in the workspace.
2. Keep recommended Turbo settings (about 8–10 steps, CFG ~1.0, Shift 3–6) unless you know you need otherwise.
3. Generate and preview the result in a few seconds on the hosted cluster.
4. Refine the prompt or settings, then export/download when satisfied.
5. Optionally call the REST API or load official weights in ComfyUI for local/self-hosted runs.

Core Features

• Z-Image-Turbo 6B — Distilled 8-step S3-DiT pipeline for fast text-to-image.
• Hosted low latency — Typical 2–4 second generations on the managed cluster.
• Bilingual prompts — Strong English and Chinese prompt and in-image text handling.
• Browser, API, ComfyUI — Same model available online, via REST, or self-hosted.
• Open weights — Apache-licensed Tongyi-MAI checkpoints for local deployment.
• Credits and teams — Stored prompts, folders, and paid volume/priority plans.
• Expanding workspace — Connected tools for related image/video creation tasks.

Use Cases

• Rapid concept art — Iterate many looks in seconds during early design reviews.
• Bilingual campaigns — Produce visuals and signage-friendly Chinese/English text.
• Pipeline integration — Call the REST endpoint or drop weights into ComfyUI workers.
• Product and lifestyle stills — Draft photoreal marketing images before a shoot.
• Local privacy runs — Self-host the open model when prompts must stay on-device.