Seedance 2.5 API

Seedance 2.5 is ByteDance’s next-generation AI video generation model, announced at the Volcano Engine FORCE conference on June 23, 2026. The headline change: a single prompt can produce a 30-second, native-4K clip in one pass — no stitching, no scene-cut splicing — with up to 50 multimodal reference inputs and more controllable generation and editing.

Seedance 2.5 vs Seedance 2.0

Capability	Seedance 2.5	Seedance 2.0 (live on HiAPI)
Max single-shot duration	30 seconds, one pass	15 seconds
Resolution	Native 4K, 10-bit color	Up to 1080p on HiAPI (4K rolling out)
Reference inputs	Up to 50 — images, audio, 3D blockouts, style refs	Up to 9 images, 3 videos, 3 audio clips
Audio	Co-generated in the same latent space (native sync)	Supported
Prompt adherence	+20% vs 2.0	Baseline
Architecture	Sparse Diffusion Transformer	Diffusion Transformer
Status	Public rollout in progress	Generally available — start here

Seedance 2.5 vs Veo 3.1, Kling 3.0, and other video models

How Seedance 2.5 stacks up against the leading AI video APIs — all of the models below are live on HiAPI today under the same API key:

Model	Max single shot	Max resolution	Audio	Reference inputs
Seedance 2.5 (coming soon)	30s, one pass	Native 4K	Native sync	Up to 50 multimodal
Seedance 2.0	15s	1080p (4K rolling out)	Yes	9 images + 3 videos + 3 audio
Veo 3.1	8s	1080p	Yes	First/last frame, I2V
Kling 3.0 Omni	15s	1080p	Yes	Multimodal, I2V
Grok Imagine	30s	720p	Yes	I2V
HappyHorse 1.1	15s	1080p	Yes	I2V / R2V

The differentiators Seedance 2.5 brings: it is the only model combining a 30-second single take with native 4K, and its 50-reference capacity is several times larger than anything else in the table. For today’s workloads, the AI Video APIs collection walks through choosing among the live models.

Seedance 2.5 release date and API availability

June 23, 2026 — ByteDance officially announced Seedance 2.5 at Volcano Engine FORCE (Beijing), alongside a Seedance 2.0 4K update and Seedream 5.0.
Now — the public rollout is underway.
HiAPI — day-one support. seedance-2.5 opens on the unified async API as soon as ByteDance releases it. No waitlist, no separate signup: an existing API key calls the new model the day it goes live, and a Seedance 2.0 integration switches over with a one-line model-ID change.

What the official showcase already proves

ByteDance’s own launch page hosts real Seedance 2.5 output. We pulled the first-screen 30-second one-take (a steampunk ornithopter sequence) and verified the file rather than quoting marketing copy:

One unbroken 30.07-second clip — no cuts or stitching seams across the full frame sequence.
Five distinct environments in a single camera move — a canyon of ancient books, brass gear fields, a mechanical horse, glass waves with a sailing ship, and a moonlit ridge.
Promo exports are 1280×720 @ 30fps (H.264/AAC) — the page serves compressed web files; full native-4K output arrives with the API.

Frame-by-frame stills and the full breakdown live in our open-source gallery: awesome-seedance-2-5-prompts.

Prompting for 30-second single takes

The jump from 15 to 30 seconds changes prompt structure more than prompt length. A 15-second prompt describes a shot; a 30-second prompt directs a scene:

Write timecoded beats — [00:00-00:08] setup → [00:08-00:20] turn → [00:20-00:30] payoff, instead of one run-on description.
Declare camera continuity explicitly — “one continuous tracking shot, no cuts” is what separates a single take from a stitched montage.
Give audio its own cues — with audio co-generated natively, lines like “score swells at 0:16, decays to a single tick” are now directable.

Ten launch-ready templates built on this structure — each with a “run it on Seedance 2.0 today” fallback — are in the prompt gallery.

What can you build with the Seedance 2.5 API?

These capabilities map to concrete production workloads:

30-second product launch videos — one-take product films with timecoded story arcs: setup, feature reveal, end card. Longer single shots remove the stitching step from ad-creative pipelines.
4K e-commerce video generation — native 4K and 10-bit color for product showcases, lifestyle scenes, and marketplace-ready assets where texture fidelity sells.
Character-consistent brand storytelling — up to 50 multimodal references lock characters, products, logos, and color grades across scenes; the practical ceiling for virtual influencers and serialized short dramas.
AI short films and story trailers — a 30-second single take covers a full scene arc, turning scripts and mood boards into cinematic concept footage.
Architecture and product previews from 3D blockouts — the new 3D white-model input drives walkthroughs and hero shots that follow exact geometry.

Preview: calling Seedance 2.5 on HiAPI

curl -X POST "https://api.hiapi.ai/v1/tasks" \
  -H "Authorization: Bearer $HIAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5",
    "input": {
      "prompt": "One unbroken 30-second tracking shot ... [00:00-00:08] ... [00:26-00:30] ...",
      "duration": 30,
      "resolution": "4k",
      "aspect_ratio": "16:9",
      "generate_audio": true
    }
  }'

The response returns a task ID; poll GET /v1/tasks/{taskId} or use a callback.url for completion — the same lifecycle as every model on HiAPI. See Unified Async API and Authentication.

Prepare today: the fastest way to be ready is to ship your integration against Seedance 2.0 now. At launch, switching is a one-line model-ID change.

Seedance 2.5 API pricing

Official Seedance 2.5 pricing has not been published yet. Confirmed rates will appear on this page and the pricing page at launch.

For budgeting today, the live Seedance family on HiAPI is the closest anchor — see current per-second rates on the pricing page. As with the rest of the Seedance family, expect billing by duration and resolution.

Available today

While Seedance 2.5 is in preparation, the Seedance family on HiAPI already covers most production workloads:

Seedance 2.0 — cinematic text-to-video, first/last-frame image-to-video, and multimodal references with optional audio.
Seedance 2.0 Fast — faster, lower-cost generation up to 720P.
Seedance 2.0 Mini — the budget tier for high-volume workloads.

FAQ

When is the Seedance 2.5 release date?

ByteDance announced Seedance 2.5 on June 23, 2026, and the public rollout is underway. HiAPI supports the model the moment it is released — watch the changelog for the launch announcement.

Is the Seedance 2.5 API available on HiAPI now?

Not yet — ByteDance has not opened API access. HiAPI’s adaptation is prepared in advance, and seedance-2.5 goes live here day-one. Until then, every workflow on this page runs on Seedance 2.0 today.

How do I get Seedance 2.5 API access?

No waitlist needed. Create an API key and integrate against Seedance 2.0 today; when 2.5 opens, the same key and the same async API work with a one-line model-ID change.

How much will the Seedance 2.5 API cost?

Official pricing has not been published. Longer clips at higher resolution generally cost more per generation, and the live Seedance family rates on the pricing page are the best budgeting anchor until confirmed 2.5 pricing lands here.

What is the difference between Seedance 2.5 and Seedance 2.0?

The official upgrades: 30-second single-shot generation (vs 15s), native 4K with 10-bit color (vs 1080p-class), up to 50 multimodal reference inputs (vs 12-class), natively co-generated audio, and +20% prompt adherence. See the comparison table above.

Can Seedance 2.5 really generate 30-second videos in one shot?

Yes — the official showcase already demonstrates it: we file-verified a 30.07-second unbroken take crossing five environments with no visible cuts. See the breakdown above.

Does Seedance 2.5 support image-to-video and reference inputs?

Yes — reference capacity is the headline upgrade: up to 50 multimodal inputs including images, audio clips, 3D blockouts, and style references. Exact API limits per media type will be documented here at launch.

How can I prepare for Seedance 2.5 API integration?

Three things you can do now: ship your pipeline on Seedance 2.0 (the async task flow is identical), organize reference assets — character sheets, product shots, style frames — for the 50-reference workflow, and structure prompts as timecoded scenes using the launch templates.