Video Generation API

Generate short videos with minimax-h3 — from a text prompt, reference images, or reference audio. Billed per second of video; see model pricing for the rate at each resolution.

Videos are asynchronous

A video takes a few minutes to generate. You submit a task, get a task ID back immediately, then poll until it finishes. Do not wait on a single request — it will time out.

1. Submit a task

POST https://jzstoken.com/api/v1/videos

curl https://jzstoken.com/api/v1/videos \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-YOUR_KEY_HERE" \
  -d '{
    "model": "minimax-h3",
    "prompt": "A paper boat drifting down a rainy city street at night, neon reflections, cinematic",
    "resolution": "480p",
    "duration": 5,
    "aspectRatio": "landscape"
  }'

Returns 202 with a task:

{
  "id": "cmuxxxxxxxxxxxxxxxx",
  "object": "video.task",
  "model": "minimax-h3",
  "status": "running",
  "resolution": "480p",
  "duration": 5,
  "cost_usd": 0.0968,
  "refunded": false,
  "created": 1790590000
}

2. Poll for the result

GET https://jzstoken.com/api/v1/videos/{id} — poll every 10–15 seconds. You can only read tasks created with your own API key.

curl https://jzstoken.com/api/v1/videos/TASK_ID \
  -H "Authorization: Bearer sk-YOUR_KEY_HERE"

running — still generating; progress is 0–100.
succeeded — data[0].url is the video. The link is temporary, download it.
failed — error.message explains why, and the charge is refunded automatically (refunded: true).

Python example

import time, requests

BASE = "https://jzstoken.com/api/v1"
HEADERS = {"Authorization": "Bearer sk-YOUR_KEY_HERE"}

task = requests.post(f"{BASE}/videos", headers=HEADERS, json={
    "model": "minimax-h3",
    "prompt": "A paper boat drifting down a rainy city street at night",
    "resolution": "480p",
    "duration": 5,
}).json()
print("submitted", task["id"], "cost $", task["cost_usd"])

while True:
    t = requests.get(f"{BASE}/videos/{task['id']}", headers=HEADERS).json()
    if t["status"] == "succeeded":
        print("video:", t["data"][0]["url"])
        break
    if t["status"] == "failed":
        print("failed:", t["error"]["message"], "(refunded)" if t["refunded"] else "")
        break
    print("progress", t.get("progress", 0), "%")
    time.sleep(10)

Parameters

Field	Description
model	Required. `minimax-h3`.
prompt	Required. Describe the scene, camera movement, timing and sound. Detailed, shot-by-shot prompts give much better results.
resolution	`480p` (default), `768p` or `1080p`. Price per second depends on this.
duration	Seconds, 1–15 (default 5). At 1080p the maximum is 10.
aspectRatio	`landscape` (default), `portrait` or `square`. Ratios such as `16:9`, `9:16` and `1:1` also work.
images	Optional. Up to 9 reference images (URLs or base64).
audios	Optional. Up to 3 reference audio clips (URLs or base64).
seed	Optional integer, for reproducible results.

Billing

Price = per-second rate for the chosen resolution × duration. The full amount is reserved from your balance when you submit, so make sure your balance covers it.
If the task fails or is blocked by content moderation, the full amount is refunded automatically.
Each task appears in your usage log with its cost; refunded tasks show a cost of $0 and the reason.