Seedance 2.5 is available through BytePlus ModelArk, Volcengine Ark and aggregators such as EvoLink. It supports 4 to 30 seconds at 480p, 720p or 1080p, subject to each API's rules. BytePlus's five-second, 16:9 text-to-video examples imply about US$0.103 to US$0.569 per second; EvoLink lists US$0.138 to US$0.739. OpenAI retired Sora 2 and the Videos API on September 24, 2026, so migration requires new requests and job handling.
Last verified: 2026-10-08.
Quick Answer
| Capability | Sora 2 API before shutdown | Seedance 2.5 official Ark API |
|---|---|---|
| Create a job | POST /v1/videos |
POST /api/v3/contents/generations/tasks |
| Prompt | prompt |
A text item in content[] |
| Duration | seconds: strings such as "4", "8", "12", "16", "20" |
duration: integer 4 to 30, or -1 for automatic |
| Output size | Pixel-string size; Pro added 1920x1080 and 1080x1920 |
resolution: 480p, 720p or 1080p; separate ratio |
| References | input_reference; reusable characters had separate IDs |
Typed image, video and audio items in content[] |
| Audio | Describe dialogue and sound in the prompt | generate_audio boolean, default true |
| Success and file | completed, then the video's content endpoint |
succeeded, then content.video_url |
OpenAI's historical video guide now explicitly marks the API unavailable. Its March 2026 parameter guide documents the later 20-second and 1080p capabilities; a migration table limited to the original 12 seconds would miss those changes.
For a browser workflow, the SeedVideo AI Seedance 2.5 page describes the Web option. Web subscriptions and provider API billing are separate offers.
TL;DR
- Choose a provider before copying model IDs or request fields.
- Official Ark uses
content,resolutionandratio; the aggregator example below usesprompt,qualityandaspect_ratio. - Set duration explicitly. The official API defaults to automatic duration, not a fixed five-second clip.
- Compare actual billable usage, including video references.
- Save the returned task ID and poll that job when a response is delayed.
- Treat Sora endpoints as historical migration inputs.
Official endpoints and a five-second request
Best for: Integrating directly with the documented BytePlus or Volcengine service.
Not for: Reusing a different provider's key, model ID or request shape.
| Service | Base URL | Model ID |
|---|---|---|
| BytePlus ModelArk | https://ark.ap-southeast.bytepluses.com/api/v3 |
dreamina-seedance-2-5-260628 |
| Volcengine Ark | https://ark.cn-beijing.volces.com/api/v3 |
doubao-seedance-2-5-260628 |
Append /contents/generations/tasks for a POST submission. Retrieve that job with GET at /contents/generations/tasks/TASK_ID. Keep the key on your server and use Authorization: Bearer YOUR_PROVIDER_API_KEY with Content-Type: application/json.
The official create-task reference requires model and content. This BytePlus example adds explicit duration and output settings:
{
"model": "dreamina-seedance-2-5-260628",
"content": [
{
"type": "text",
"text": "A paper sailboat drifts across a quiet pond at sunrise. The camera tracks slowly beside it, with warm natural light, soft reflections, gentle ripples and one continuous shot."
}
],
"duration": 5,
"resolution": "480p",
"ratio": "16:9",
"generate_audio": true,
"watermark": false
}
For Volcengine, use its base URL, its model ID and a key issued for that service. The official defaults are duration: -1, resolution: "720p", ratio: "adaptive", audio enabled and watermark disabled. Omitting duration lets the model choose within its supported range, which changes the budget.
The BytePlus create-task reference documents the structured request fields and defaults.
The retrieve-task reference documents the result. A create response containing an ID means the job was accepted; read the final status before using its file URL.
The aggregator request is a different contract
Best for: Applications using the EvoLink account, API gateway and billing system.
Not for: Assuming its field names, model aliases or prices are universal ByteDance API rules.
EvoLink's text-to-video documentation uses POST https://api.evolink.ai/v1/videos/generations, with model seedance-2.5-text-to-video. Its required fields are model and prompt. The matching explicit settings are duration: 5, quality: "480p", aspect_ratio: "16:9" and generate_audio: true.
Do not send Ark's content array unchanged to that route. Image, reference, editing and extension tasks have their own model IDs and allowed fields. For example, image-to-video takes one or two image_urls, while reference-to-video accepts supported image, video or audio arrays.
The provider's English API reference shows its own text-to-video request and task response.
Poll GET https://api.evolink.ai/v1/tasks/TASK_ID using the returned ID. The documented terminal statuses are completed and failed; completed URL-producing jobs expose results. Official Ark instead uses succeeded and content.video_url. Centralize that status and result mapping in your provider adapter.
Pricing by resolution and billing unit
Official list prices
The following are public standard rates, not private-account discounts. The BytePlus price page quotes USD; the Volcengine price page quotes CNY.
| Provider / input | 480p or 720p, per million video tokens | 1080p, per million video tokens | Verified |
|---|---|---|---|
| BytePlus, no video input | US$10.70 | US$11.70 | 2026-10-08 |
| BytePlus, with video input | US$6.40 | US$7.00 | 2026-10-08 |
| Volcengine, no video input | CNY 70 | CNY 77 | 2026-10-08 |
| Volcengine, with video input | CNY 42 | CNY 46 | 2026-10-08 |
These are generated-video tokens, not the number of words in your prompt. Final cost follows the applicable rate multiplied by usage.completion_tokens, divided by 1,000,000. Video input changes the accounting and has a minimum token floor.
For the official five-second, 16:9 examples with no video input:
| Output | BytePlus estimated clip cost | Approximate cost per output second | Volcengine estimated clip cost |
|---|---|---|---|
| 480p | US$0.514 | US$0.103 | CNY 3.36 |
| 720p | US$1.156 | US$0.231 | CNY 7.56 |
| 1080p | US$2.843 | US$0.569 | CNY 18.71 |
The per-second column divides those specific clip estimates by five. It is not a universal fixed rate for every aspect ratio, reference input or mode. No currency conversion is used in this comparison.
BytePlus lists separate Seedance 2.5 rates for resolution and the presence of video input.
Public aggregator quote
EvoLink's public pricing table shows these rounded USD rates:
| Mode | 480p per billable second | 720p per billable second | 1080p per billable second | Verified |
|---|---|---|---|---|
| Text/image generation or references without video | US$0.138 | US$0.296 | US$0.739 | 2026-10-08 |
| Video references, editing or extension | US$0.084 | US$0.180 | US$0.450 | 2026-10-08 |
Without video input, billable duration is the output duration. With video input, it is output duration plus the larger of total input-video duration and output duration. A five-second input with an eight-second output therefore bills 16 seconds: at 720p, approximately US$2.88. A five-second 720p text-to-video job is approximately US$1.48 at the displayed rounded rate.
The public price table distinguishes output-only billing from video-reference billing.
Account quotes can differ, and precise credits may produce slightly different rounded USD totals. Draft generation and conversion are billed separately; optional features have their own charges. Use the final job's accounting for reconciliation. The video-tool cost comparison covers other pricing approaches.
Limits, access and safe retries
The official Seedance 2.5 guide lists up to 30 reference images, 10 reference videos and 10 audio clips. Video-reference duration totals at most 30 seconds; audio-reference duration also totals at most 30 seconds. Mode-specific restrictions still apply. Editing, for example, uses adaptive framing and automatic duration.
Published official account classes have these ceilings:
| Account class | Creation requests per minute | Processing concurrency |
|---|---|---|
| Individual | 180 | 3 |
| Enterprise | 600 | 10 |
These limits apply to the same primary account and model. They do not prove that a particular account has model access. The official guide lists retrieval at 20 queries per second. Keep polling within the shared limit, and distinguish queued work from an HTTP 429 rate-limit response. No fixed aggregator concurrency or Seedance 2.5-specific free API grant is established by the public sources cited here.
Access has its own requirements. The official activation guidance lists a balance above US$30 for BytePlus or above CNY 200 for Volcengine, a qualifying savings plan at the stated tier, or an eligible Seedance resource pack with remaining quota. Confirm the applicable account and region route before funding it. Real-human portrait inputs need the documented trusted-asset process; do not assume ordinary image URLs are accepted for that case.
For asynchronous jobs, the practical workflow is:
- Validate the selected provider's body and required asset inputs locally.
- Submit once and save the returned task ID with your own job record.
- Poll that ID with backoff; follow Retry-After when the provider returns it.
- On a timeout or transport error, check the existing task before creating another chargeable job.
- Read final status and usage, then save the file while its URL is valid.
Both reviewed API families document result URLs valid for 24 hours. Ark also documents seven-day task retention. Expired output links need different handling from a failed generation.
For output-quality context, the Seedance benchmark overview summarizes third-party results; it does not establish your application's throughput.
Migrating a Sora project
Best for: Reusing your prompt library, job queue and application-level workflow.
Not for: Keeping the retired Sora endpoint or assuming a base-URL change completes migration.
Use the OpenAI deprecation notice as the shutdown source. Map prompt, seconds, size and references into the new provider's schema. Replace Sora's completed status and content-download call with the chosen provider's result contract. Rebuild reference assets in the target service; Sora character IDs are not a documented portable asset format.
Start with one short scene and explicit settings before extending the migration. This reusable prompt can also help compare the browser workflow:
A paper sailboat drifts across a quiet pond at sunrise. The camera tracks slowly beside it, with warm natural light, soft reflections, gentle ripples and one continuous shot.
Try this prompt in SeedVideo AI
If you prefer an interactive Web workflow, use SeedVideo AI's pricing to evaluate that separate offer. The Sora alternatives overview covers additional choices.
FAQ
Is there a free Seedance 2.5 API allowance?
A specific free Seedance 2.5 generation grant was not established in the cited public material. General cloud free-token offers and free account registration are not proof that this video model is included. Check the provider's activation and billing conditions.
Do I need enterprise verification?
Requirements depend on provider, region and workflow. Published individual quotas do not guarantee approval. Volcengine's advanced creation-rights guidance explicitly requires enterprise verification for that product; do not apply that requirement to every ordinary call or automatically copy it to BytePlus.
How many videos can run concurrently?
The official guide lists 3 processing jobs for individual accounts and 10 for enterprise accounts. Account access and granted limits still apply. The aggregator's public Seedance 2.5 material does not establish one universal concurrency allowance.
Does the API generate audio?
Official Ark and the reviewed aggregator route document generate_audio, default true. Set it false for silent output. Audio-reference requirements and permitted combinations depend on the mode you select.
Are API outputs watermark-free?
Official Ark documents watermark with a default of false. That does not establish every other provider's export policy. The aggregator material reviewed here does not confirm a Seedance 2.5-specific watermark guarantee, and its allowed fields should be followed exactly.
What must change in a Sora application?
Change authentication, host, model ID, body fields, reference-asset handling, status mapping and output retrieval. Preserve your own job IDs and retry controls. Sora 2 and its Videos API have been unavailable since September 24, 2026.
Choose the integration you need
Use the API when your application needs automated jobs and result handling. For an interactive workflow, open the SeedVideo AI text-to-video generator. More related data is in the AI video pricing collection.
