Seedance 2.5 Flash
Generate videos with Seedance 2.5 Flash using text, first and last frames, or multimodal references. This page covers the complete request-to-result workflow for this model.
API model ID: seedance-2-5-flash
Generation is asynchronous. Save the taskId returned when creating a task, then query its status or receive a webhook.
Capabilities
| Feature | Supported values |
|---|---|
| Output resolution | 480p · 720p |
| Output duration | 4–30 seconds |
| Aspect ratio | 16:9, 4:3, 1:1, 3:4, 9:16, 21:9, adaptive |
| Reference images | Up to 30 images |
| Reference videos | Up to 10 videos |
| Reference audio files | Up to 10 audio files |
| All references combined | Up to 50 reference files in total |
| Total duration per video/audio group | 30 seconds |
| seed | -1 to 2147483647 |
Pricing & credits
Video generation is charged in credits based on the billable duration in seconds. Without video input, the billable duration is the output duration; with video input, it also includes the reference video duration.
The table below lists the credits charged per second, not the total cost of a task. The rate depends on the model, output resolution, and whether reference videos are supplied in reference-to-video mode. See the formulas and examples below the table for the total cost calculation.
| Output resolution | Without video input | With video input |
|---|---|---|
480p | 7 credits/second | 5 credits/second |
720p | 10 credits/second | 7 credits/second |
- Without video input: output seconds × no-video rate.
- With video input: (output seconds + measured reference-video seconds) × with-video rate. The server measures the total reference-video duration and rounds it up to whole seconds before billing.
- Image or audio references alone use the no-video rate. The reference-video billing rate applies only in reference-to-video mode when video references are supplied.
Billing examples
5-second 720p text-to-video: 5 × 10 = 50 credits.
5-second 720p output with a 5-second reference video: (5 + 5) × 7 = 70 credits.
Credits are reserved on submission and charged on success. Failed or timed-out tasks enter the refund flow. A refund_failed billing status means the refund did not complete; check the API logs or contact support.
How credits are charged when duration=-1
When duration is set to -1, the final output duration is not fixed; the model determines it.
In most cases, set duration to the video length you actually need instead of -1. We recommend using -1 only for video editing, not for other video generation scenarios.
| Reference inputs | How the charge is calculated | Example |
|---|---|---|
| With reference videos | Add up the durations of all reference videos and round the total up to the next whole second, calling it T. The charge is (T + T) × the with-video rate: one T represents the estimated output duration, and the other represents the input video duration. | 720p with a 5-second reference video: (5 + 5) × 7 = 70 credits. |
| Without reference videos (images or audio only) | Use 30 seconds as the estimated output duration. The charge is 30 × the without-video rate. | 720p without reference videos: 30 × 10 = 300 credits. |
Authentication
Create an API key in the dashboard. The complete key is shown only once. Keep it on your server and send it as a Bearer token on every request.
Base URL
https://api.seevio.aiAuthorization: Bearer sk_live_your_api_key
Content-Type: application/jsonSet the SEEVIO_API_KEY environment variable before running these examples. JavaScript examples run on your server with Node.js; Python examples use the requests package.
Quickstart
Submit this minimal request, save the returned taskId, then use the task query example below. The credits value in the create response is the reserved amount.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "A cat surfing at sunset, cinematic lighting",
"duration": 5,
"resolution": "720p",
"seed": 42,
"generation_type": "text-to-video",
"aspect_ratio": "16:9",
"generate_audio": true
}
}'Create task response example
After the request above is accepted, the API returns this JSON response. taskId is the task identifier used for subsequent status queries; credits is the number of credits reserved for this task. This response confirms task creation, not that the video is ready. You need to poll the task status or use a Webhook to receive the video results.
{
"taskId": "3f2aK9mR7xQp4TnZ8bLc6YwH",
"credits": 50
}Create a task
POST https://api.seevio.ai/v1/videos/generationsSend a JSON object containing model and input, plus an optional callback_url. Always specify the model ID shown on this page; omitting model selects seedance-2-0.
Request body
| Field | Type | Required | Description & constraints |
|---|---|---|---|
model | string | Yes | Model ID. To use Seedance 2.5 Flash, set this field to seedance-2-5-flash. |
callback_url | string | No | Public HTTPS endpoint for completion and failure POST callbacks. Private networks and localhost are not allowed. Example: https://example.com/webhooks/seevio |
input | object | Yes | Generation settings. Must contain a non-empty prompt. |
Input parameters
Provide image_urls, video_urls and audio_urls as arrays of URL strings (string[]). Every supplied URL must be publicly accessible over HTTPS, including media ignored by the selected mode.
| Field | Type | Required | Default | Description & constraints |
|---|---|---|---|---|
input.prompt | string | Yes | — | A prompt is required in every mode. It can contain at most 10,000 characters before trimming and cannot consist entirely of whitespace. Example: A cat surfing at sunset |
input.generation_type | string | No | text-to-video | Supported values text-to-video | image-to-video | reference-to-videotext-to-video, image-to-video or reference-to-video. If omitted, defaults to text-to-video and uses only the prompt. image-to-video uses 1–2 ordered frame images. reference-to-video uses images, videos and/or audio; audio-only references are supported with a required prompt. |
input.image_urls | string[] | Conditional | [] | Supply image URLs to guide appearance. In image-to-video, provide 1 image for the first frame or 2 in first-frame, last-frame order. In reference-to-video, provide at most 30 images. Each image must be smaller than 30 MB; see the media requirements below for formats and dimensions. All reference types combined allow at most 50 files. Omitted defaults to []; ignored in text-to-video. Example: ["https://example.com/first-frame.jpg"] |
input.video_urls | string[] | Conditional | [] | Supply reference video URLs to guide motion or edit/extend a video, only in reference-to-video. At most 10 files, each up to 100 MB and 2–30 seconds long; their combined duration must not exceed 30 seconds. All reference types combined allow at most 50 files. The server measures video duration for billing. Omitted defaults to []; ignored in other modes. Example: ["https://example.com/source.mp4"] |
input.audio_urls | string[] | Conditional | [] | Supply reference audio URLs to guide sound, only in reference-to-video. At most 10 MP3/WAV files, each up to 15 MB and 2–30 seconds long; their combined duration must not exceed 30 seconds. All reference types combined allow at most 50 files. Audio can be the only reference type; prompt remains required. Omitted defaults to []; ignored in other modes. Example: ["https://example.com/music.mp3"] |
input.duration | integer | No | 5 | Supported values -1 | 4–30Integer output duration from 4 to 30 seconds. Also accepts -1 only in reference-to-video. Use it with a source video for editing; billing follows the special rule above. Example: 5 |
input.aspect_ratio | string | No | adaptive | Supported values 16:9, 4:3, 1:1, 3:4, 9:16, 21:9, adaptive16:9, 4:3, 1:1, 3:4, 9:16, 21:9 or adaptive. If omitted, defaults to adaptive, which lets the model determine the output ratio. For video editing or extension, use adaptive. image-to-video only accepts adaptive; omit this field or set it to adaptive. Example: adaptive |
input.resolution | string | No | 720p | Supported values 480p | 720p480p or 720p. If omitted, defaults to 720p in every supported generation mode. Higher resolutions are rejected. Example: 720p |
input.generate_audio | boolean | No | true | Supported values true | falsetrue requests synchronized audio; false disables audio generation. If omitted, defaults to true. Applies to all supported generation modes. Example: true |
input.watermark | boolean | No | false | Supported values true | falsetrue requests an AI watermark on the output; false requests output without that watermark. If omitted, defaults to false. Applies to all supported generation modes. Example: false |
input.web_search | boolean | No | false | Supported values true | falsetrue allows web search when available for the model; false disables it. If omitted, defaults to false. Enabling it does not guarantee that a search will be performed. Example: false |
input.return_last_frame | boolean | No | false | Supported values true | falsetrue requests the generated video’s final frame; false does not request it. If omitted, defaults to false. Applies to all supported generation modes. When requested and available, the URL is returned as data.last_frame_url; otherwise it is null. Example: true |
input.seed | integer | No | — | Supported values -1 to 2147483647Integer from -1 to 2147483647. -1 chooses a random seed. Example: 42 |
input.output_format | string | No | mp4 | Supported values mp4 | movmp4 or mov. If omitted, defaults to mp4. Selects the output container format in all supported generation modes. |
Boolean fields must be JSON true or false, not strings or numbers.
Create response
HTTP 200 returns taskId (string) and credits (number). This acknowledges task creation, not completion. The amount below corresponds to the 5-second 720p quickstart.
{
"taskId": "3f2aK9mR7xQp4TnZ8bLc6YwH",
"credits": 50
}Generation modes & examples
Text to video
Generate from a text prompt. Media URLs are not forwarded in this mode.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "A cat surfing at sunset, cinematic lighting",
"duration": 5,
"resolution": "720p",
"seed": 42,
"generation_type": "text-to-video",
"aspect_ratio": "16:9",
"generate_audio": true
}
}'First frame
Provide one image as the first frame, then describe the motion in your prompt.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "A cat surfing at sunset, cinematic lighting",
"duration": 5,
"resolution": "720p",
"seed": 42,
"generation_type": "image-to-video",
"image_urls": [
"https://example.com/first-frame.jpg"
],
"aspect_ratio": "adaptive"
}
}'First and last frames
Provide two image URLs in order: first frame, then last frame. This example also requests the generated video's last frame.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "A cat surfing at sunset, cinematic lighting",
"duration": 5,
"resolution": "720p",
"seed": 42,
"generation_type": "image-to-video",
"image_urls": [
"https://example.com/first-frame.jpg",
"https://example.com/last-frame.jpg"
],
"aspect_ratio": "adaptive",
"return_last_frame": true
}
}'Multimodal reference
Combine image, video and audio references. The prompt remains required. Reference-video input changes the billing formula.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "Follow the reference camera movement and keep the character consistent. Use the audio for ambience.",
"duration": 5,
"resolution": "720p",
"seed": 42,
"generation_type": "reference-to-video",
"image_urls": [
"https://example.com/character.jpg"
],
"video_urls": [
"https://example.com/camera.mp4"
],
"audio_urls": [
"https://example.com/ambience.mp3"
],
"aspect_ratio": "adaptive"
}
}'Audio reference
Use audio as the only reference type, with a required text prompt describing the desired video.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "Create a coastal sunrise scene matching the rhythm of this audio.",
"duration": 5,
"resolution": "720p",
"seed": 42,
"generation_type": "reference-to-video",
"audio_urls": [
"https://example.com/music.mp3"
]
}
}'Video editing
Describe the edit and provide the source video. Set duration=-1 and use adaptive aspect ratio. Use a source clip of at least 4 seconds for this workflow. The billing rule for duration=-1 is shown above.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "Edit the source video: change the character’s coat to blue and preserve the camera movement.",
"duration": -1,
"resolution": "720p",
"seed": 42,
"generation_type": "reference-to-video",
"video_urls": [
"https://example.com/source.mp4"
],
"aspect_ratio": "adaptive"
}
}'Video extension
Describe how the source video should continue. Use adaptive aspect ratio and set a normal output duration within the model's range.
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "Continue the camera movement from the source video, revealing a forest clearing.",
"duration": 8,
"resolution": "720p",
"seed": 42,
"generation_type": "reference-to-video",
"video_urls": [
"https://example.com/source.mp4"
],
"aspect_ratio": "adaptive"
}
}'Query a task
GET https://api.seevio.ai/v1/tasks/{taskId}Replace the example ID with the taskId returned by creation. Queries only return tasks owned by the API key's user; inaccessible or unknown IDs return HTTP 404.
Poll every 10–20 seconds as a starting point, back off on HTTP 429 and stop when the status is completed or failed. Prefer webhooks for production. Each code example below performs one query.
curl --fail-with-body https://api.seevio.ai/v1/tasks/3f2aK9mR7xQp4TnZ8bLc6YwH \
-H "Authorization: Bearer $SEEVIO_API_KEY"| Status | Description & constraints |
|---|---|
queued | Accepted and waiting for submission. |
generating | Generation is in progress. |
completed | Terminal success. Download data.results before expiry. |
failed | Terminal failure. Inspect failed_reason and billing_status. |
| Field | Type | Description & constraints |
|---|---|---|
id | string | Task identifier. This is taskId from the create response. |
created_at | number | Task creation time as Unix seconds. |
model | string | The public model ID used for this task. |
billing_status | string | reserved, charged, refunded or refund_failed. |
credits | number | Credits reserved for this task. This value is retained after a refund; inspect billing_status to determine the billing outcome. |
failed_reason | string | null | Failure reason on failed tasks; null otherwise. Failed query responses omit data. |
data | object | Present on non-failed task queries. Contains output and processing details. |
data.results | string[] | Video URL array. Empty until completion or after the video's expiry. |
data.video_expires_at | string | null | Video expiry as an ISO 8601 timestamp, or null before it is available. Save the result before this time. |
data.last_frame_url | string | null | Last-frame URL when requested and available, otherwise null. |
data.processing_time | number | null | Task processing duration in seconds when available, otherwise null. |
Completed task: query response with video results
When the query returns status=completed, video generation has finished. Read the video URLs from data.results and download them before data.video_expires_at. billing_status=charged indicates that the reserved credits have been charged.
{
"id": "3f2aK9mR7xQp4TnZ8bLc6YwH",
"created_at": 1788652800,
"model": "seedance-2-5-flash",
"is_draft": false,
"draft_task_id": null,
"status": "completed",
"billing_status": "charged",
"credits": 50,
"failed_reason": null,
"data": {
"results": [
"https://cdn.seevio.ai/api/videos/example.mp4"
],
"video_expires_at": "2026-09-07T00:00:00Z",
"last_frame_url": null,
"processing_time": 48
}
}Failed task: query response with failure and billing details
When the query returns status=failed, generation has ended unsuccessfully. Read failed_reason for the cause and billing_status for the refund outcome. In this example, refunded means the credits were returned. credits retains the original reserved amount, and the response does not include data.
{
"id": "3f2aK9mR7xQp4TnZ8bLc6YwH",
"created_at": 1788652800,
"model": "seedance-2-5-flash",
"is_draft": false,
"draft_task_id": null,
"status": "failed",
"billing_status": "refunded",
"credits": 50,
"failed_reason": "provider_failed"
}Webhooks
Set callback_url in the create request to receive a JSON POST when the task completes or fails. Return a 2xx response within 15 seconds. Failed deliveries are retried; process repeated deliveries idempotently by task ID.
Your callback endpoint must accept POST requests with a JSON request body (Content-Type: application/json).
Create a task with a callback
curl --fail-with-body https://api.seevio.ai/v1/videos/generations \
-H "Authorization: Bearer $SEEVIO_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "seedance-2-5-flash",
"input": {
"prompt": "A cat surfing at sunset, cinematic lighting",
"duration": 5,
"resolution": "720p",
"seed": 42,
"generation_type": "text-to-video",
"aspect_ratio": "16:9",
"generate_audio": true
},
"callback_url": "https://example.com/webhooks/seevio"
}'Task completed: successful callback payload
When generation succeeds, the callback contains status=completed. Use id to identify the task and data.results to retrieve the video URLs. Download and save the results before data.video_expires_at.
created_at is the event creation time; task_created_at is the task creation time in Unix seconds. Examples show the recommended fields; responses may contain additional fields.
{
"id": "3f2aK9mR7xQp4TnZ8bLc6YwH",
"created_at": 1788652848,
"model": "seedance-2-5-flash",
"status": "completed",
"task_created_at": 1788652800,
"billing_status": "charged",
"credits": 50,
"failed_reason": null,
"data": {
"results": [
"https://cdn.seevio.ai/api/videos/example.mp4"
],
"video_expires_at": "2026-09-07T00:00:00Z",
"last_frame_url": null,
"processing_time": 48
}
}Task failed: failure callback payload
When generation fails, the callback contains status=failed. Use id to identify the task, failed_reason for the failure reason and credits_refunded for the number of credits refunded.
{
"id": "3f2aK9mR7xQp4TnZ8bLc6YwH",
"created_at": 1788652848,
"model": "seedance-2-5-flash",
"status": "failed",
"task_created_at": 1788652800,
"billing_status": "refunded",
"credits": 50,
"failed_reason": "provider_failed",
"credits_refunded": 50
}Receiver example
export async function POST(request: Request) {
const callbackData = await request.json();
if (callbackData.status === "completed") {
const videoUrls = callbackData.data.results;
// Save the video URLs and mark this task as completed in your application.
console.log(callbackData.id, videoUrls);
}
if (callbackData.status === "failed") {
const { failed_reason, credits_refunded } = callbackData;
// Record the failure reason and refunded credits for this task.
console.error(callbackData.id, failed_reason, credits_refunded);
}
return new Response(null, { status: 200 });
}This Next.js example reads the JSON callback body and handles completed and failed tasks directly. Add persistence and task-ID deduplication for your application; queue slow work before acknowledging the callback.
Media requirements & limitations
- All media and callback URLs must be public HTTPS URLs. Avoid localhost, private IPs and files that require cookies or login. Reference video/audio URLs must resolve to directly readable media.
- In reference-to-video, provide at least one reference, with no more than 30 images, 10 videos, 10 audio files and 50 materials combined. Total video duration and total audio duration must each be at most 30 seconds.
- text-to-video ignores all media references. image-to-video forwards only the first/last-frame images and ignores video and audio references. Use reference-to-video to combine media.
- Each reference video and audio file must be 2–30 seconds. For video editing examples, use source clips of at least 4 seconds.
Image requirements
- Each image must be smaller than 30 MB.
- Supported formats: jpeg, png, webp, bmp, tiff, gif.
- Aspect ratio (width ÷ height): 0.4 to 2.5, inclusive.
- Width and height must each be between 300 and 6,000 pixels, inclusive.
Video requirements
- Supported formats: mp4, mov.
- Each video must not exceed 100 MB.
- Frame rate: 24 to 60 FPS, inclusive.
- Aspect ratio (width ÷ height): 0.4 to 2.5, inclusive.
- Total pixels (width × height): 407,696 to 8,295,044, inclusive. For example, 614 × 664 = 407,696 and 3,326 × 2,494 = 8,295,044. These are examples of pixel counts, not fixed width and height requirements.
Audio requirements
- Supported formats: wav, mp3.
- Each audio file must not exceed 15 MB.