Community

Seedance Video Generation API

Call Seedance (Volcengine Ark) video generation through LMU AI with the native async protocol: create, query, list and cancel tasks, plus a full polling example.

Seedance (Volcengine Ark) video generation uses the vendor's native asynchronous protocol: replace the address in the official SDK or curl command with the LMU AI address and the key with your LMU AI API key, and the request body stays unchanged. For the shared rules on authentication, billing and watermarks, concurrency limits, and error codes, see the Video Generation API overview.

Base URL

https://api.lmuai.ai
OperationEndpoint
Create a taskPOST /api/v3/contents/generations/tasks
Query a taskGET /api/v3/contents/generations/tasks/{id}
List tasksGET /api/v3/contents/generations/tasks
Cancel / deleteDELETE /api/v3/contents/generations/tasks/{id}

Video generation must be enabled for your group

The API key's group must have "Video generation" enabled; otherwise calls return 403. The model name must match the name enabled for your group exactly (case-sensitive); check available models with GET /v1/models.

1. Create a task

POST /api/v3/contents/generations/tasks

ParameterTypeRequiredDescription
modelstringYesModel name; must exactly match the name enabled for your group (case-sensitive)
contentarrayYesInput content. Text: {"type":"text","text":"…"}; reference image: {"type":"image_url","image_url":{"url":"…"},"role":"first_frame"}, where role can be first_frame (first frame) or reference_image (reference image)
resolutionstringNoResolution tier: 480p / 720p / 1080p
durationintegerNoVideo length in seconds
ratiostringNoAspect ratio: 16:9 / 9:16 / 1:1 / 4:3
generate_audiobooleanNoWhether to generate audio as well
curl https://api.lmuai.ai/api/v3/contents/generations/tasks \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5",
    "content": [
      {"type": "text", "text": "Waves gently lapping the beach at dusk, slow motion"}
    ],
    "resolution": "720p",
    "duration": 5,
    "ratio": "16:9",
    "generate_audio": true
  }'
import requests

BASE = "https://api.lmuai.ai"
HEADERS = {"Authorization": "Bearer YOUR_API_KEY"}

resp = requests.post(f"{BASE}/api/v3/contents/generations/tasks", headers=HEADERS, json={
    "model": "seedance-2.5",
    "content": [{"type": "text", "text": "Waves gently lapping the beach at dusk, slow motion"}],
    "resolution": "720p",
    "duration": 5,
    "ratio": "16:9",
    "generate_audio": True,
})
resp.raise_for_status()
task_id = resp.json()["id"]
print(task_id)
const BASE = "https://api.lmuai.ai";
const headers = {
  Authorization: "Bearer YOUR_API_KEY",
  "Content-Type": "application/json",
};

const resp = await fetch(`${BASE}/api/v3/contents/generations/tasks`, {
  method: "POST",
  headers,
  body: JSON.stringify({
    model: "seedance-2.5",
    content: [{ type: "text", text: "Waves gently lapping the beach at dusk, slow motion" }],
    resolution: "720p",
    duration: 5,
    ratio: "16:9",
    generate_audio: true,
  }),
});
const data = await resp.json();
console.log(data.id);

Response:

{
  "id": "cgt-20261022103211-ab12c"
}

To use a first-frame reference image, append an image item to content:

{
  "model": "seedance-2.5",
  "content": [
    {"type": "text", "text": "The camera slowly pushes in as waves gently lap the beach"},
    {"type": "image_url", "image_url": {"url": "https://example.com/first-frame.png"}, "role": "first_frame"}
  ],
  "resolution": "720p",
  "duration": 5
}

Notes

  • In first-frame mode (role is first_frame), do not send ratio: the aspect ratio follows the first-frame image, and the upstream rejects the request with 400 if you send it.
  • callback_url is not forwarded upstream; you will not receive a callback even if you pass it. Use polling instead.

2. Query a task

GET /api/v3/contents/generations/tasks/{id}

status values: queued → running → succeeded / failed / cancelled. On success, the video URL is in content.video_url.

curl https://api.lmuai.ai/api/v3/contents/generations/tasks/TASK_ID \
  -H "Authorization: Bearer YOUR_API_KEY"

Response (success):

{
  "id": "cgt-20261022103211-ab12c",
  "model": "seedance-2.5",
  "status": "succeeded",
  "content": {
    "video_url": "https://…/video.mp4"
  },
  "usage": {
    "completion_tokens": 108900,
    "total_tokens": 108900
  },
  "created_at": 1792629131,
  "updated_at": 1792629262
}

3. List tasks

GET /api/v3/contents/generations/tasks?limit=20&offset=0

Returns the tasks created with the current API key (limit from 1 to 100, default 20). The response is {object, data[], has_more}. Task records remain queryable after the video link expires.

4. Cancel / delete

DELETE /api/v3/contents/generations/tasks/{id}

  • Tasks still in the queue (queued) can be cancelled at no charge.
  • Tasks that have started generating (running) cannot be cancelled and return 409.
  • Finished tasks can have their upstream record deleted, with no refund.

5. Complete example: create and poll

import time, requests

BASE = "https://api.lmuai.ai"
HEADERS = {"Authorization": "Bearer YOUR_API_KEY"}

# 1. Create the task
resp = requests.post(f"{BASE}/api/v3/contents/generations/tasks", headers=HEADERS, json={
    "model": "seedance-2.5",
    "content": [{"type": "text", "text": "Waves gently lapping the beach at dusk, slow motion"}],
    "resolution": "720p", "duration": 5, "ratio": "16:9",
})
resp.raise_for_status()
task_id = resp.json()["id"]

# 2. Poll until a final state (every 10 seconds)
while True:
    task = requests.get(f"{BASE}/api/v3/contents/generations/tasks/{task_id}", headers=HEADERS).json()
    status = task["status"]
    if status in ("succeeded", "failed", "cancelled", "canceled", "expired"):
        break
    time.sleep(10)

# 3. Download (the link is valid for 24 hours)
if status == "succeeded":
    url = task["content"]["video_url"]
    with open(f"{task_id}.mp4", "wb") as f:
        f.write(requests.get(url).content)
else:
    print("Generation failed:", task)

6. Billing and error codes

Seedance is billed by output tokens. An estimated amount is frozen when the task is created, and an insufficient balance returns 402; on success the task is settled by actual usage, and failed and cancelled tasks are fully refunded. Each user and each API key can have at most 5 tasks generating at the same time; beyond that, requests return 429.

For the full rules, see Billing and watermarks, Concurrency limits and Error codes in the overview. For the other provider, see the Wan Video Generation API.

Last updated:

Still have questions? Ask in the LMU AI Community (in Chinese) and sign in with your console account.

On this page

1. Create a taskPOST /api/v3/contents/generations/tasks2. Query a taskGET /api/v3/contents/generations/tasks/{id}3. List tasksGET /api/v3/contents/generations/tasks?limit=20&offset=04. Cancel / deleteDELETE /api/v3/contents/generations/tasks/{id}5. Complete example: create and poll6. Billing and error codes