Z-Image Turbo

z-image-turboOur GPUsText to image

Tongyi-MAI's 8-step distilled Z-Image Turbo on VelvetAPI's own GPUs: fast, photorealistic text to image.

Inputs

Send these in the task's input object. Only required fields are needed; GET /v1/models/z-image-turbo returns them as a JSON Schema.

Field	Type	Default	Details
`prompt`required	string	—	What to draw. Z-Image Turbo reads the first ~512 tokens (about 2,000 characters of English) and ignores the rest. up to 20,000 characters
`width`	integer	—	Multiple of 16, 512-2048. Defaults to 1024. 512 to 2,048 · multiple of 16
`height`	integer	—	Multiple of 16, 512-2048. Defaults to 1024. 512 to 2,048 · multiple of 16
`steps`	integer	—	Sampling steps (default 8). 4 to 16
`seed`	integer	—	0 to 2,147,483,647
`num_images`	integer	`1`	1 to 4
`output_format`	string	`"jpeg"`	one of jpeg, png, webp

Example

Create a task with your API key, then poll it or pass a webhook_url. user is your own id for the end user who asked for the image.

curl https://api.velvetapi.dev/v1/tasks \
  -H "Authorization: Bearer $VELVET_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "z-image-turbo",
  "user": "user-123",
  "input": {
    "prompt": "A candid photograph of a woman with auburn hair reading on a sunlit balcony, soft morning light, 35mm film grain."
  }
}'

import { VelvetAPI } from "@velvetapi/sdk";

const velvet = new VelvetAPI({ apiKey: process.env.VELVET_API_KEY });

const task = await velvet.tasks.create({
  model: "z-image-turbo",
  user: "user-123",
  input: {
    prompt: "A candid photograph of a woman with auburn hair reading on a sunlit balcony, soft morning light, 35mm film grain.",
  },
});
const done = await velvet.tasks.wait(task.id);
console.log(done.state, done.outputs.map((output) => output.url));

import os

from velvetapi import VelvetAPI

client = VelvetAPI(api_key=os.environ["VELVET_API_KEY"])

task = client.tasks.create(
    model="z-image-turbo",
    user="user-123",
    input={
        "prompt": "A candid photograph of a woman with auburn hair reading on a sunlit balcony, soft morning light, 35mm film grain.",
    },
)
done = client.tasks.wait(task["id"])
print(done["state"], [output["url"] for output in done["outputs"]])

Response 202 Accepted

{
  "id": "0b9c6c2e-4d3f-4a51-9a8e-2f7d1c3b5a90",
  "object": "task",
  "model": "z-image-turbo",
  "state": "queued",
  "user": "user-123",
  "price_usd": 0.006,
  "outputs": [],
  "error": null,
  "meta": {},
  "created_at": "2026-10-20T06:31:00.000Z",
  "started_at": null,
  "finished_at": null
}

Next: tasks and webhooks, errors and the API reference.