The first integration decision is the route: use doubao-seedance-2-5-flash with /ai/v1/videos, then keep the task ID through polling and download. The current Flash schema lists only that native async route. Changing the model name in an older /v1/videos client is not enough.
A November 16, 2026 check confirmed the schema and a rejection, not a generated clip. Both the short ID and doubao-seedance-2-5-flash-261115 returned a schema and rejected 1080p with 400 schema_violation. A 4-second 480p creation attempt instead returned 402 insufficient_quota, so this article has no successful generation, latency measurement, or end-to-end MOV/edit test.
Start with the schema and account prerequisites
The Flash model page uses the short ID in these examples. Enable asynchronous media tasks in the AIHubMix console, check access and account credit, and export AIHUBMIX_API_KEY. The documented activation error is 403 async_not_enabled; the measured quota error was a separate 402 insufficient_quota.
Retrieve the public schema before deployment:
curl --fail-with-body -sS \
'https://aihubmix.com/call/schema/models/doubao-seedance-2-5-flash/endpoints' \
| jq '.endpoints[] | select(.path == "/ai/v1/videos") | .request.schema'
As checked on November 16, output is 480p or 720p, defaulting to 720p. Duration accepts integers 4–30 or -1 for model-selected length. Explicit duration makes the requested budget easier to estimate.
Submit once, then follow the existing task
The Video Generation API has three stages: create with POST /ai/v1/videos, query with GET /ai/v1/videos/{id}, and download from GET /ai/v1/videos/{id}/content. A successful creation response is acceptance; it is not completion. A status query can return HTTP 200 and still report status: "failed".
Install requests with python3 -m pip install requests. Save this as seedance_flash.py and run python3 seedance_flash.py from the shell holding your key. It requests a text-only 10-second 720p 16:9 clip with audio, without depending on externally hosted reference media. The code follows the schema and documented lifecycle; the quota limitation prevented a full test.
import os
import time
from pathlib import Path
import requests
BASE_URL = "https://aihubmix.com"
HEADERS = {
"Authorization": f"Bearer {os.environ['AIHUBMIX_API_KEY']}",
"Content-Type": "application/json",
}
PAYLOAD = {
"model": "doubao-seedance-2-5-flash",
"prompt": "A matte ceramic espresso cup sits on wet slate as warm morning light sweeps across rising steam. The camera slowly dollies in. Quiet cafe ambience.",
"duration": 10,
"aspect_ratio": "16:9",
"resolution": "720p",
"generate_audio": True,
}
# Submit once. A network timeout does not prove that creation failed.
response = requests.post(
f"{BASE_URL}/ai/v1/videos", headers=HEADERS, json=PAYLOAD, timeout=60
)
response.raise_for_status()
task = response.json()
task_id = task["id"]
print(f"Task ID: {task_id}", flush=True)
deadline = time.monotonic() + 20 * 60
while task["status"] not in {"completed", "failed", "cancelled"}:
if time.monotonic() >= deadline:
raise TimeoutError(f"Still waiting for {task_id}; query it again, do not recreate it.")
time.sleep(15)
response = requests.get(
f"{BASE_URL}/ai/v1/videos/{task_id}", headers=HEADERS, timeout=30
)
response.raise_for_status()
task = response.json()
if task["status"] != "completed":
raise RuntimeError(f"Task {task_id}: {task.get('error') or task['status']}")
response = requests.get(
f"{BASE_URL}/ai/v1/videos/{task_id}/content", headers=HEADERS, timeout=120
)
response.raise_for_status()
Path("seedance_flash_10s.mp4").write_bytes(response.content)
print("Saved seedance_flash_10s.mp4")
The 20-minute deadline only bounds this client's wait. It neither cancels the server task nor promises a completion time. Keep the printed ID if querying or downloading fails. Retry the query or download against that task; do not automatically create a second generation after a timeout or an uncertain provider state. Active polling should use the video detail route, which refreshes provider state, rather than the unified task list.
What the rejection actually proves
Both tested IDs returned HTTP 400 when the request used resolution: "1080p". These were the returned error fields; the response's trace ID is omitted:
{
"error": {
"code": "schema_violation",
"message": "`resolution` must be one of: 480p, 720p, null.",
"type": "invalid_request_error"
}
}
That is an observed gateway validation result. The other invalid combinations below were read from the current schema, not separately submitted:
| Constraint | Accepted configuration |
|---|---|
| Output resolution | 480p or 720p |
| Reference caps | 10 images / 5 videos / 5 audio |
| Reference arrays | One non-empty array |
| Frame ratio | Adaptive |
| Last frame | First frame also present |
For media guidance, input_references entries contain type (image_url, video_url, or audio_url) and your accessible media url. First/last-frame workflows use frame_images, with frame_type and nested image_url.url. Do not combine the two non-empty arrays.
Add optional capabilities with their limits in view
The Async Tasks documentation describes webhook_url callbacks. Flash exposes that field but currently omits webhook_events_filter and rejects unspecified properties, so omit the filter when following this schema. The documented default delivers all three terminal events. Delivery is at least once: deduplicate on event_id, and confirm final state with a task query. Task-level callbacks have no separate signing secret; the docs describe account-level subscriptions when signature verification is required.
For a real person's likeness, follow the Doubao Real-Person Assets Guide: create an asset group, create a personal verification session, have that person complete web verification, register the asset, wait for active status, and use asset://<ASSET_ID>. The documented doubao_real_person_required denotes Doubao's specific privacy rejection; it was not reproduced here.
The Volcengine family guide documents MOV, editing, and extension. Flash's gateway schema exposes extra.output_format (mp4/mov) and extra.omni_reference_task_type (auto, reference, edit, extend), but their end-to-end behavior was not verified in this check. Explicit task type requires non-empty references. Edit also requires a video reference, adaptive ratio, and duration: -1; extend requires a video reference and adaptive ratio.
Ark documents InvalidParameter.TaskTypeConstraint for inferred-type parameter conflicts and InvalidParameter.TaskTypeMismatch when the model disagrees with an explicit type. No exact AIHubMix mapping was measured. Read the task error instead of assuming a particular gateway code. Flash has no Draft mode, and the current AIHubMix schemas for Flash and flagship omit draft and draft_task.
There is also a documentation ambiguity: extra.priority, extra.safety_identifier, and extra.tools are exposed but described as “2.0 series only,” while Ark lists priority and tools for the 2.5 family too. Their presence does not establish their effect through AIHubMix.
Budget and save the result
At official BytePlus online prices, Flash is $7.00/M tokens without video input and $4.30/M with it at 480p/720p. The sample's estimate is 10 × 1280 × 720 × 24 / 1024 = 216,000 tokens, or $1.51, versus $2.31 on flagship 2.5 at 720p. These are computed BytePlus costs, not a measured AIHubMix bill. Video input adds input duration and can trigger a minimum token floor, so the lower token rate does not ensure a lower total bill.
The gateway documents a 32 MiB request-body limit (413 request_too_large). Public media URLs avoid Base64 expansion in the body, while the referenced files still need to meet model limits. Download completed videos promptly. The existing 2.5 guide warns of roughly 30-minute URLs, but Flash's exact retention was not verified; expires_at: null does not imply permanent storage. Documented errors include 410 artifact_expired and 429 too_many_downloads.
Start with the current schema and the text-only client, keep the task ID, and verify generation on an account with sufficient credit before depending on optional capabilities.
Keep reading: the Seedance 2.5 Flash series
- Choosing the tier before integrating it? Seedance 2.5 Flash vs 2.5 and 2.0: Check Your Output Requirements First
Sources
Checked November 16, 2026; validation responses recorded at 08:26 UTC.
- Video Generation API (AIHubMix Docs)
- Async Tasks API (AIHubMix Docs)
- Doubao Real-Person Assets Guide (AIHubMix Docs)
- ModelArk Pricing (BytePlus)
- Seedance 2.5 Tutorial (Volcengine Ark)
Top comments (0)