Seedance 2.5 API: release date, pricing and cost per clip

Seedance 2.5 costs $0.137/s at 480p, $0.215/s at 720p and $0.549/s at 1080p through API Stock when you supply no input video. Ten seconds at 720p cost $2.15. With a reference video, a different rate applies to input-video seconds plus requested output seconds, so the cheaper per-second rate does not automatically mean a cheaper request.

Updated 15 October 2026. At 720p, output without input video now costs $0.215/s: ten seconds cost $2.15. With reference video, $0.129/s applies to input plus output seconds. The tables below use the checked catalog snapshot; the original cover retains the prices shown on publication, 30 September 2026.

When was Seedance 2.5 released?

ByteDance published its Seedance 2.5 announcement on 31 July 2026. The announcement describes longer single-pass generation and expanded multimodal referencing. That release date is distinct from availability or feature support on any particular API service.

On API Stock, seedance-2.5 accepts 4–30 seconds and offers 480p, 720p and 1080p. Those are the options in this service's current schema; an announced capability elsewhere does not establish that our endpoint exposes it. Local region/time-range editing is not currently available through this endpoint.

How much does a clip cost without input video?

A text-only request, a first-frame image request and an image/audio reference request all use the regular output rate. Sound is on by default; the catalog does not charge a separate audio surcharge.

The balance debit retains fractional cents. For example, five seconds at 720p cost $1.075; a cent-denominated display may show $1.08. Twenty such requests cost $21.50. Budget from the tariff or precise quote, not by multiplying a rounded display.

If you omit optional settings, the five-second 480p default costs $0.685. Four seconds at 480p cost $0.548 and are the minimum-duration request; the default is not the cheapest possible request.

How is a reference-video or extension request billed?

The service measures reference-video duration before charging. Its billing rule is:

request cost = (total input-video seconds + requested output seconds) × input-video tariff, with fractional cents retained in the balance debit.

At 720p, a ten-second reference video plus ten requested output seconds cost (10 + 10) × $0.129 = $2.58. Without the video, ten output seconds cost $2.15. Images and audio alone do not trigger the input-video tariff. For several videos, the service adds their measured durations and rounds the total up to whole seconds for billing, rather than using just the longest one. A 10.04-second reference counts as eleven input seconds: with ten output seconds at 720p that is 21 × $0.129 = $2.709.

These are billing examples, not claims that an extension preserves a scene perfectly. Inspect the output and include unsuccessful creative attempts in your budget. The model reference lists the input limits and supported modes.

How does Seedance 2.5 compare with the cheaper tiers?

For a ten-second 720p request with no input video:

2.5 supports up to thirty images across its image fields, ten video references and ten audio references, within the duration limits in the reference. The 2.0 models accept fewer references and have a fifteen-second output limit. That makes 2.5 relevant for longer single takes or a larger reference brief; it does not prove that every short clip will look better.

For a simple ten-second shot, Mini's $0.38 tariff total is about 5.7 times cheaper than 2.5's $2.15. Check whether its output meets your brief before moving to a more expensive tier. If you need 4K, Seedance 2.0 has that option; 2.5 on this endpoint does not. A new render at another resolution is a new take, not an upscale or a guaranteed match to a fixed seed.

What does a working Seedance 2.5 request look like?

This text-only request produces a ten-second 720p task, priced at $2.15 in the snapshot:

curl -X POST https://api.api-stock.com/api/v1/generation/create \
  -H "Authorization: Bearer $API_STOCK_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5",
    "input": {
      "prompt": "A locked-off studio shot of one red running shoe on a pale grey table. A band of sunlight moves slowly across it. The shoe stays still. Sound: quiet room tone, no music.",
      "duration": 10,
      "resolution": "720p",
      "size": "16:9",
      "generateAudio": true
    }
  }'

The response carries a taskId; poll task status or provide a top-level webhook. For a first-frame request, supply your own accessible image using imageWithRoles and role: "first_frame", and leave size at adaptive. This adds an image, not input-video seconds. The image-to-video guide explains the workflow; the task docs explain completion and refunds.

This request is a recipe, not a published Seedance test result. For actual first-take Grok/Kling clips and their limitations, see the video cost comparison. Use compare to inspect two models' parameters without assuming identical behaviour.

FAQ

When was Seedance 2.5 released?

ByteDance's official announcement is dated 31 July 2026. Availability and exposed features depend on the service; API Stock currently lists it as seedance-2.5.

How much does ten seconds at 720p cost?

$2.15 without input video at the 15 October 2026 snapshot. Ten input-video seconds plus ten output seconds cost $2.58 under the separate input-video tariff.

What is the cheapest request?

Four seconds at 480p with no input video cost $0.548. The five-second 480p default is $0.685.

Does a first-frame image trigger reference-video billing?

No. Images and audio alone retain the regular output tariff. Input-video duration triggers the input-video rate on both measured input-video seconds and requested output seconds.

Can Seedance 2.5 make 4K or edit a region of a video here?

The current endpoint offers 480p, 720p and 1080p, not 4K. Reference generation and extension are supported; local region/time-range editing is not exposed here.

Is Seedance 2.5 better than Mini for short clips?

This article establishes differences in tariffs, duration and inputs, not a quality ranking. Compare results on your own brief and account for every paid attempt.