Seedance 2.5 available now in ChatSlide

Seedance 2.5 API

ByteDance Seedance 2.5 generates up to 30 seconds of video in a single pass, takes up to 50 multimodal references, and co-generates audio with the picture.

Call it directly, or generate with it inside ChatSlide today — pick Seedance 2.5 per video, see the credit cost before you press go, and skip the API key and the polling loop.

Quick answer

The Seedance 2.5 API is ByteDance's video generation endpoint, served through Volcano Engine and authorized providers. It produces 4–30 second clips in one pass at 480p or 720p, with 1080p live from 17 August 2026 and no 4K tier, accepts text, image, video, and audio references — up to 50 multimodal inputs — and co-generates audio. Calls are asynchronous: submit a job, then poll or take a webhook. Billing is per token, not per second: a 5-second 1080p clip costs ¥18.71 (¥3.74/s) and a 30-second one about ¥112, roughly US$16.6. Only successful generations are charged.

Seedance 2.5 specifications

Max duration4–30 seconds, generated in a single pass (no stitching, no extension calls)
Resolution480p and 720p; 1080p from 17 August 2026. Seedance 2.5 has no 4K tier on Volcano Engine — Seedance 2.0 does. 4K on 2.5 means an upscale step, not native generation
Aspect ratios21:9, 16:9, 4:3, 1:1, 3:4, 9:16, 9:21
Reference inputsMarketed as up to 50 multimodal references; published schemas cap them per type (≈20 images, ≈6 video clips, ≈6 audio files)
AudioCo-generated with the picture via a generate_audio flag — not dubbed on afterwards
Camera control3D camera blockouts, plus a camera_fixed flag to lock the shot
ModesText-to-video, image-to-video, reference, keyframe (first/last frame), video edit and extend
Determinismseed parameter (0–4294967295, or -1 for random)
Call patternAsynchronous — submit, receive a job id, poll or take a webhook. ~2 minutes typical for a short clip

Compiled from published provider schemas and Volcano Engine documentation, August 2026. Limits move — verify against your provider's current docs before you build against them.

The 50-reference number needs a caveat

“Up to 50 references” is the marketed ceiling for total multimodal assets. The published request schemas are stricter per type: roughly 20 reference images, 6 video clips, and 6 audio files in a single call. If you plan a shot around 40 character stills, you will hit a per-type cap long before 50.

Practically, this rewards curation over volume: a handful of well-chosen, clearly-roled references conditions the shot better and costs less than a large unlabelled pile.

How the API call works

Every provider serving Seedance 2.5 uses the same asynchronous shape: submit a job, get an id, then poll or take a webhook. Endpoint paths and parameter names differ, so treat this as the pattern rather than a copy-paste target.

# 1. submit
curl -X POST https://<provider>/v2/videos/generate \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "slow dolly in on a ceramic espresso cup, morning light, steam rising",
    "duration": 10,
    "resolution": "1080p",
    "generate_audio": true,
    "reference_image_urls": ["https://.../cup-front.jpg"],
    "seed": 42
  }'
# -> 202 Accepted { "jobId": "..." }

# 2. poll until done
curl https://<provider>/v2/jobs/$JOB_ID -H "Authorization: Bearer $API_KEY"

Short clips typically complete in about two minutes. Budget longer for 30-second and upscaled jobs, and handle failed generations — per-second billing normally only charges for successful ones.

Seedance 2.5 API pricing

ByteDance does publish official Seedance 2.5 rates — on Volcano Engine's Chinese model-pricing page, which is why most English guides still say pricing is unavailable and extrapolate from Seedance 2.0. The extrapolations are wrong in a structural way: Seedance 2.5 is billed by token, not per second.

These are ByteDance's own worked examples for a 5-second 16:9 clip with no video input — the only figures it states directly in currency.

Resolution5s clipPer secondPer second (USD)30s clip
480p¥3.36¥0.67$0.099¥20.16 (~$2.99)
720p¥7.56¥1.51$0.224¥45.36 (~$6.72)
1080p¥18.71¥3.74$0.554¥112.20 (~$16.63)

The number worth carrying away: a 30-second 1080p clip — the capability Seedance 2.5 is sold on — costs about ¥112 (US$16.6) at list price, or roughly ¥81 (US$12.0) during the promotional window. That is about 33× a five-second 480p generation, so the headline feature is also the expensive one.

The token rates behind those numbers

480p / 720p, no video input¥70.00 per million tokens
480p / 720p, with video input¥42.00 per million tokens
1080p, no video input¥77.00 list — ¥55.44 during the promo
1080p, with video input¥46.00 list — ¥33.12 during the promo

Token usage is estimated as (input duration + output duration) × width × height × frame rate / 1024, and the price is that figure multiplied by the token rate. Actual billing uses the usage.completion_tokens field returned by the API, so treat the formula as a budgeting tool rather than an invoice.

Three things the formula does not show. Video input raises the floor, not just the rate: a 5-second 1080p output with 2–30 seconds of input video runs ¥20.12–¥78.25, and there is a minimum token charge that applies when input includes video. Only successful generations are billed — clips that fail moderation are free. And 1080p is discounted 28% until 17 September 2026 (from 14 August), which is the only reason the per-second floor reaches ~¥2.7.

Read from Volcano Engine's official 模型价格 documentation on 17 August 2026 (UTC+8). USD converted at 1 USD = 6.7487 CNY. Promotional rates expire — verify before committing to volume.

Seedance 2.5 vs Seedance 2.0

FeatureSeedance 2.5Seedance 2.0
Max single-shot length30s4–15s
Reference inputsUp to 50 multimodalNot quantified
Resolution ceiling1080p (from 17 Aug 2026); no 4K tier480p–4K by variant
AudioCo-generatedLimited
Camera control3D blockoutsPrompt-based
Official list price, 5s 1080p¥18.71 (¥3.74/s)¥12.39 (¥2.48/s)
Official list price, 5s 720p¥7.56 (¥1.51/s)¥4.97 (¥0.99/s)

Seedance 2.5 vs Veo 3.1 vs Sora 2

FeatureSeedance 2.5Veo 3.1Sora 2
Max single-shot length30s8s~20s
Resolution ceiling1080p (2.5); 4K on 2.01080p1080p
Reference conditioningUp to 50 inputsLimitedLimited
AudioCo-generatedCo-generatedCo-generated
Camera control3D blockoutsPrompt-basedPrompt-based
BillingPer token, not per secondPer second of outputPer generation

Vendor-published specs and official Volcano Engine pricing, current as of 17 August 2026.

Prompting a 30-second single take

Block the shot, then describe it
Seedance 2.5 responds to camera language — "slow dolly in, then orbit left 90°" holds better than "cinematic camera movement". Write the move, not the mood.
One subject beat per 5 seconds
A 30-second single take still needs a beat structure. Give roughly one action change per five seconds so the model has something to resolve toward instead of drifting.
Reference by role, not by pile
Label what each reference is for — character, product, style plate, environment. Dumping 40 unlabelled images costs more and conditions less than eight labelled ones.
Lock the seed before you tune
Set a fixed seed while you iterate on the prompt, so you are measuring prompt changes rather than sampling noise. Release the seed only for final variations.

If you do not want to integrate anything

Most people searching for a video API do not want an API — they want a finished video. Raw model access means you still own prompt engineering, reference curation, job polling, retries on failed generations, storage, captioning, and voiceover.

Upload a deck, document, or script — no prompt engineering required
AI avatars and voice cloning in 17 languages
Narration, captions, and export handled end to end
No API key, no polling loop, no per-second invoice

Seedance 2.5 is live in ChatSlide — what a clip costs

You pick Seedance 2.5 per video, choose the resolution and length, and see the exact credit cost before you generate. No API key, no polling loop, and no token arithmetic: the raw API bills per token with a minimum-token floor and a multiplier when your input includes video, so the price of a clip is only knowable after the fact.

Resolution5-second clip30-second clipPlan
480p30 credits180 creditsPro
720p55 credits330 creditsPro
1080p140 credits840 creditsUltimate

Text-to-video and image-to-video, 4–30 seconds, 16:9 / 9:16 / 1:1 and the other native ratios, audio generated with the picture. Failed generations are not charged. See credit pricing for how credits map to plans.

ChatSlide runs generative video as a finished pipeline rather than a raw endpoint — Seedance 2.5 sits next to Veo 3.1 in the same picker, and the output lands in your library rather than on an expiring CDN URL you have to go and fetch. Building at team scale? Talk to us about Enterprise for SSO, team billing, and shared templates.

Frequently Asked Questions

What is the Seedance 2.5 API?
Seedance 2.5 is ByteDance's video generation model, served through Volcano Engine and authorized third-party providers. It generates 4 to 30 seconds of video in a single pass, accepts up to 50 multimodal references (images, video clips, audio), outputs 480p to 1080p natively with 4K upscaling, and co-generates audio with the picture.
How much does the Seedance 2.5 API cost?
ByteDance publishes official rates on Volcano Engine, and it bills Seedance 2.5 by token rather than per second — which is why English guides extrapolating a per-second rate from Seedance 2.0 get it wrong. Its own worked examples for a 5-second 16:9 clip with no video input are ¥3.36 at 480p (¥0.67/s), ¥7.56 at 720p (¥1.51/s), and ¥18.71 at 1080p (¥3.74/s). A 30-second 1080p clip therefore costs about ¥112, roughly US$16.6. Token rates are ¥70 per million tokens at 480p/720p without video input and ¥42 with it; 1080p is ¥77 and ¥46 at list, discounted 28% until 17 September 2026. Only successful generations are billed.
Can Seedance 2.5 really take 50 reference inputs?
That is the marketed ceiling for total multimodal references. Published API schemas are stricter per type — roughly 20 reference images, 6 video clips, and 6 audio files in one request. Design for the per-type caps, not the headline number.
Is Seedance 2.5 native 4K?
No, and it is not native 1080p either until 17 August 2026, when ByteDance switched that tier on. Volcano Engine lists no 4K price for Seedance 2.5 at all — but it does for Seedance 2.0, at ¥26 per million tokens (¥25.27 for a 5-second clip). So if you specifically need 4K out of ByteDance today, the older model is the one that generates it; 4K on 2.5 means an upscale step in your pipeline, which changes both quality and cost.
What changed from Seedance 2.0?
Length is the headline: 30 seconds in one generation instead of 4 to 15 second clips that had to be stitched or extended. It also raises the reference ceiling, adds 3D camera blockouts for directed moves, and improves audio co-generation. In practice the payoff is consistency — character, product, and lighting hold across a full shot.
How do I call it?
Generation is asynchronous everywhere it is served. You POST a prompt plus parameters, get a job id back, then poll the job endpoint or register a webhook. Short clips typically finish in around two minutes; 30-second jobs at high resolution take longer, so build for polling rather than a blocking request.
Do I need to write code to use Seedance 2.5?
Only if you want raw model access. Seedance 2.5 is available in ChatSlide today: pick it per video, choose 480p, 720p or 1080p and 4 to 30 seconds, and the credit cost is shown before you generate — 30 credits for a 5-second 480p clip, 140 for 5 seconds at 1080p, 840 for the full 30 seconds at 1080p. Seedance is on the Pro plan and 1080p on Ultimate. Text-to-video and image-to-video both work, audio is generated with the picture, failed generations are not charged, and the finished file is saved to your library instead of an expiring CDN link. No API key, no polling loop, and no token arithmetic.
Is Seedance 2.5 better than Veo or Sora?
For long single takes and heavy reference conditioning, Seedance 2.5 leads on published specs; on resolution all three top out at 1080p, and 4K out of ByteDance means Seedance 2.0 or an upscale step. For eight-second social clips the gap narrows and availability matters more. And if the real job is turning slides or a script into a presentable video, the pipeline around the model matters more than the model.

Generate with Seedance 2.5 in ChatSlide

Create a 4–30 second clip from a prompt and references, with synchronized audio. Free to start, no credit card.

Get Started Free