Skip to main content

Overview

Seedance 2.0 Fast is a ByteDance video generation model for fast text-to-video, first-and-last-frame animation, and multimodal reference workflows. Use this page when you are ready to call the API after trying the model in the APIXO playground.

Endpoint and authentication

Base URL:
All requests require your APIXO API key:
Submit requests also require:

Copy-paste async quickstart

This minimal request submits a text-to-video task and returns a taskId.
Successful response:
Save the taskId; you need it to poll for the final result.

Poll for result

Processing response:
Success response:
Failed response:
Parse resultJson after state becomes success:

Request body

Text-to-video

First-and-last-frames

Omni-reference

Parameters

string
default:"async"
required
Result delivery mode. Use async for polling with statusTask, or callback for webhook delivery.
string
Required when request_type is callback. Must be a public HTTPS URL that can receive the final task payload. See Webhooks.
object
required
Seedance 2.0 Fast input parameters.

Mode-specific constraints

  • text-to-video uses the prompt and generation settings only.
  • first_and_last_frames requires image_urls with 1-2 images. The first image is the starting frame; the second image, when present, is the ending frame.
  • omni_reference can combine image_urls, video_urls, and audio_urls.
  • omni_reference requests with only audio_urls, and no images or videos, are rejected.
  • Reference media must be publicly reachable URLs so the API can fetch or probe them.

Response format

Submit task response

POST /generateTask/seedance-2-0-fast returns a task ID when the task is accepted:
integer
API status code. 200 means the task was accepted.
string
Human-readable status message.
string
Unique task identifier used with the status endpoint.

Status response fields

string
Unique task identifier.
string
Current task state: pending, processing, success, or failed.
string
JSON string containing the generated video URLs in resultUrls. Present when state is success.
string
Machine-readable failure code. Present when state is failed.
string
Human-readable failure message. Present when state is failed.
integer
Task creation timestamp in Unix milliseconds.
integer
Task completion timestamp in Unix milliseconds. Present after completion.
integer
Processing duration in milliseconds. Present after completion when timing data is available.
object
Final billing snapshot under data.billing, returned on successful terminal status and callback responses for new tasks. Older tasks may omit this object; see Billing for compatibility rules.
string
Settlement status. The final snapshot uses final.
string
Opaque identifier of the billing rule captured when this task was created. Do not infer the rule from polling time.
number
Actual total reference-video duration in seconds. Zero without video references; this is not raised to the minimum billable input.
number
For new tasks with video references, ceil(outputVideoSeconds * 2 / 3), rounded upward to a whole second. Zero without video references or under the original rule.
number
The larger of inputVideoSeconds and minimumInputVideoSeconds; zero without video references.
number
Output duration used for billing, in seconds.
number
Total billed seconds: outputVideoSeconds + billableInputVideoSeconds.
string
Duration unit. The value is second.

Webhook callback mode

Use callback mode when your backend should receive the final result automatically instead of polling.
Successful terminal callbacks for new tasks include the same data.billing snapshot as status polling. See Webhooks for delivery requirements and retry behavior.

Billing

New non-LLM requests use 1 credit = $0.001. Calculate the full amount, including billable quantity and add-ons, apply the applicable account discount, then round the request total upward to the next whole credit. Agreed fixed unit prices retain their precision and replace the standard price and discount; final rounding still applies. See credit conversion and rounding. Seedance 2.0 Fast is billed per second. The selected resolution and whether omni_reference includes video_urls determine the unit price. Billing formulas:
  • Without video reference: output duration seconds x per-second rate
  • With video reference: (output duration seconds + max(actual total reference video seconds, ceil(output duration seconds * 2 / 3))) x per-second rate
ceil rounds upward to the next whole second. The minimum applies once to the total reference-video duration, not separately to each video. For example, 5 output seconds with 3 actual reference seconds bills 5 + max(3, 4) = 9 seconds; 6 output seconds with 2 actual reference seconds bills 10 seconds. This rule applies to tasks created after the billing update. Existing tasks retain the rule captured at creation: with video references, (output duration + actual total reference video duration) * unit price; without video references, output duration only. Existing tasks retain their captured unit prices, discounts, and settlement policy. Polling or receiving a callback after the update does not change an existing task’s rule. text-to-video and first_and_last_frames always use the “without video reference” rate. In omni_reference mode, “with video reference” applies only when video_urls is provided. Image and audio references do not add extra billable seconds. Successful terminal status and callback responses for new tasks include data.billing. Duration values use seconds with three decimal places; minimumInputVideoSeconds is a whole-second ceiling. inputVideoSeconds remains the actual input duration, while billableInputVideoSeconds is the charged input duration. The snapshot remains stable on repeated polling. Older tasks may omit billing or the new fields; clients must not require them for historical tasks. For example, the billing snapshot for 5 output seconds with 3 reference-video seconds includes:
For current API rates in USD, see API pricing.

Latency and polling

Actual latency may vary by prompt complexity, reference media, provider route, and current queue load.
For production workloads, use callback mode to reduce polling overhead during longer video tasks.
No visible status change during the first few minutes can be normal for this model. Keep polling with backoff instead of immediately retrying the same task.
Result URLs can be temporary. Download generated videos promptly after task completion.

Errors and troubleshooting

HTTP errors

Common request validation errors

Task failure codes

See Error Codes for the full error reference.

MCP integration

The complete model schema declares result decoding in mcp.result and supported file bindings in mcp.attachments. These are client integration metadata, not generation input parameters. Attachment paths: image_urls, video_urls, audio_urls. Preserve the user’s file order. Model parameter validation remains on the API.