Seedance 2.5 Launch Status: 30 Seconds Confirmed, 4K Unverified

Seedance 2.5 is now callable on reAPI as doubao-seedance-2.5-face, with 480p, 720p and 1080p output and published production prices. As of August 17, 2026, the live route supports 4–30-second generation, up to 30 images, 10 videos, and 10 audio references, plus the editing and extension workflows described in the API reference.^[7]^[8] Native 4K and named Maya or Blender integrations are still not part of the documented 2.5 API.

That distinction still matters because many Seedance 2.5 summaries combine capabilities shown by ByteDance, figures repeated from launch coverage, and features that belong to Seedance 2.0. This guide now separates the live reAPI contract from claims that remain outside it, then compares the documented model with MiniMax H3.

TL;DR

Confirmed by ByteDance: one-pass 30-second video, more multimodal reference input, second-level scene control, deeper video editing, and multilingual output demonstrations.^[1]
The live reference ceiling: up to 30 images, 10 videos, and 10 audio clips; 50 materials in total.^[8]
Still not part of the documented 2.5 route: native 4K, 10-bit output, and direct Maya or Blender integration. reAPI currently accepts 480p, 720p and (since August 2026) 1080p — 4K is still not served.^[8]
Public API access is live on reAPI: the model ID is doubao-seedance-2.5-face.^[7]^[8]
Current pricing: $0.118589/s at 480p and $0.266824/s at 720p without source video; $0.071153/s and $0.160094/s with source video.^[7]

Seedance 2.5 specs: confirmed versus unverified

Seedance 2.5 confirmed features compared with unverified launch claims

Claim	Status on August 17, 2026	What the evidence says
30-second continuous generation	Officially confirmed	ByteDance's ModelArk promotion says “30s continuous output” and shows a segmented 30-second example.^[1]
Expanded multimodal references	Documented on reAPI	Up to 30 images, 10 videos, and 10 audio clips per request.^[8]
Timestamp or segment control	Officially demonstrated	Prompts assign actions to time ranges inside a continuous clip.^[1]
Video editing	Officially demonstrated	One example removes background people while preserving the main subject.^[1]
Multilingual speech and lip sync	Officially demonstrated	The product page includes a multilingual performance example.^[1]
Native 4K in Seedance 2.5	Not supported on reAPI	The live 2.5 route accepts 480p, 720p and (since August 2026) 1080p; 4K remains documented for Seedance 2.0.^[2]^[8]
30 images, 10 videos, 10 audio inputs	Documented on reAPI	These are the current per-request ceilings for `doubao-seedance-2.5-face`.^[8]
50 total references	Documented on reAPI	The three media limits add up to 50 reference assets.^[8]
Maya and Blender integration	Unverified	No named DCC plug-in, import format, or integration guide appears in the first-party material checked.
Public API availability	Live on reAPI	Use `doubao-seedance-2.5-face`; 480p, 720p and 1080p rates are published on the model page.^[7]^[8]

The table separates what the live reAPI contract accepts from claims that remain outside that contract. Production forms and estimates should follow the documented route, not a launch summary.

What 30-second generation changes in practice

Moving from Seedance 2.0's documented 15-second ceiling to a 30-second single pass is more than doubling a number.^[4] It reduces the number of seams in any sequence longer than one shot.

A 60-second product film built from 15-second generations needs at least four clips and three joins. At 30 seconds, the theoretical minimum falls to two clips and one join. Fewer joins mean fewer opportunities for wardrobe, lighting, geometry, voice, and screen direction to drift.

The tradeoff is failure cost. A bad detail near second 27 can make a longer generation expensive to discard. The best workflow will probably remain modular:

Block the story into five- to ten-second beats.
Describe those beats with explicit time ranges.
Generate a 30-second take when continuity is more valuable than independent rerolls.
Keep difficult dialogue, product text, or transformation shots separate when precision matters more.

Pricing is now concrete. Without source video, a 30-second 720p request is 8,005 credits, or $8.005 after rounding. Use 30 seconds when continuity justifies the larger reroll cost, not simply because the maximum exists.^[7]

The multimodal reference system is the real upgrade

Longer clips attract the headline, but the reference system is more important for repeatable production. The official Seedance 2.5 example addresses eight images individually and gives each one a role inside the scene.^[1] That suggests a workflow where references are not merely blended into a style average; they can be assigned to characters, objects, environments, and beats.

The useful way to prepare is to build a reference manifest now:

Reference role	Recommended asset	What it should control
Character identity	Clean front and three-quarter portraits	Face, hair, age, and wardrobe anchors
Product identity	Neutral product packshot	Shape, materials, color, and logo placement
Environment	Wide establishing frame	Architecture, palette, weather, and time of day
Motion	Short source clip	Gesture, blocking, camera path, or physical timing
Audio	Clean speech or sound reference	Voice, rhythm, ambience, or lip-sync target

The 30/10/10 split is now documented on reAPI, but the role-based manifest remains useful. It keeps a request understandable and makes it easier to remove references that compete with one another.^[8]

Timestamp control turns a prompt into a compact edit timeline

ByteDance's demonstration divides a continuous generation into timed segments. That makes a Seedance prompt closer to a shot list:

0–6s: Wide establishing shot. The actor enters from frame left.
6–14s: Track backward at walking speed. Keep the product in the right hand.
14–22s: Medium close-up. The actor stops and delivers one short line.
22–30s: Slow orbit to the product. Hold the final composition for two seconds.

This is not proof of frame-exact editing. “Second-level control” describes the interface demonstrated by ByteDance; it does not guarantee that every action starts on the requested frame.^[1] Prompts should still leave physical transitions enough time to complete.

The editing demo is also notable. Removing everyone except one protagonist from an existing clip is a different task from text-to-video. reAPI now documents video edit and video extend as Seedance 2.5 task types, alongside the reference limits, 480p/720p/1080p resolutions, and billing rules.^[8]

Native 4K is the wrong claim to lead with

Several summaries describe Seedance 2.5 as a native 4K launch. The production evidence says otherwise: reAPI accepts 480p, 720p and (since August 2026) 1080p for 2.5, while ByteDance documents 4K output for Seedance 2.0.^[2]^[8]

Do not send 1080p or 4k to the current 2.5 route. If the contract changes later, the live model page and request schema should replace this dated comparison.

For production, resolution is only one part of quality. Temporal stability, facial continuity, typography, motion coherence, and sound synchronization usually determine whether a shot survives review. Evaluate those at the delivery resolution you actually need rather than treating 4K as a proxy for reliability.

Is there a Maya or Blender integration?

Not in the official material currently available. Some coverage interprets 3D-oriented demonstrations as Maya or Blender integration. A model accepting a render, viewport capture, depth-like image, or turntable video is not the same as a native plug-in that reads scene geometry, cameras, materials, or animation curves.

A real DCC integration announcement should identify at least one of these:

a supported Maya or Blender plug-in;
a scene or interchange format such as USD, Alembic, or FBX;
preserved camera and object metadata;
an official installation guide and version requirements.

Until that exists, the safe description is 3D-reference-friendly video generation, not Maya/Blender integration.

Seedance 2.5 vs MiniMax H3

MiniMax H3 remains useful as a comparison because both models now have callable routes and documented constraints. MiniMax launched H3 on July 31 with a public video API guide, a named model value, limits, resolution, native stereo audio, and pricing.^[5]^[6]

Capability	Seedance 2.5	MiniMax H3 / Hailuo 03
Maximum documented duration	30 seconds in official product demo	15 seconds in public API documentation
Resolution	480p or 720p on reAPI	Up to 2K
Native audio	Multilingual audio/lip-sync demonstrated	Native stereo audio documented
Mixed references	Up to 30 images, 10 videos, and 10 audio clips	Up to 9 images, 3 videos, and 3 audio clips
Timeline control	Second-level segmented prompting demonstrated	Prompt-directed timing; no comparable timestamp interface documented at launch
Editing existing video	Demonstrated	Reference-video generation documented; selective editing is not the launch headline
Public API contract	Published on reAPI	Published

Seedance 2.5 has the longer continuous-duration and editing workflow, while MiniMax H3 reaches 2K. Both can now be inspected and called through a defined API. Read the dedicated MiniMax H3 vs Seedance 2.5 comparison for the model decision, or the Hailuo H3 API guide for implementation details.

Realistic human avatars—and the Hollywood question

The demonstrations point toward longer, more controllable human performances: multiple characters, timed actions, speech, lip sync, and continuity across a scene. That makes advertising, virtual presenters, music videos, previsualization, and synthetic performers obvious test cases.

It does not establish Hollywood adoption. Studios evaluate rights provenance, consent, union terms, editability, security, insurance, and repeatability alongside visual quality. A polished launch reel answers only the last of those, and only on selected examples.

My prediction is narrower: the first durable “synthetic actors” will not be one-prompt replacements for human performers. They will be licensed digital characters with locked identity packs, approved voices, reference manifests, and human-controlled shot boundaries. Seedance 2.5's multimodal direction fits that pipeline, but adoption depends on the rights and control layer around the model.

What active Seedance users should do now

There is useful migration work now that the model ID and rates are known:

Route by deliverable. Keep Seedance 2.0 for 1080p/4K or lower-cost tiers; test Seedance 2.5 for 16–30-second takes and video edits.
Create a fixed evaluation set. Include faces, hands, products, dialogue, camera moves, text, and a difficult 20–30-second narrative.
Label every reference by role. Identity, environment, motion, composition, and audio should be separable.
Test timestamped prompts on 2.5. Compare the requested event boundaries with the rendered result.
Abstract the model version in code. Keep task submission, polling, storage, and review independent from the model string.
Validate against the live schema. Enforce 480p/720p, the documented reference limits, and the source-video billing rule before updating a production form.

The Seedance 2.5 model page now publishes live prices, and the API documentation identifies the callable model as doubao-seedance-2.5-face.

Frequently asked questions

Has Seedance 2.5 launched?

Yes. As of August 17, 2026, reAPI exposes doubao-seedance-2.5-face with live prices and a public parameter reference.^[7]^[8]

Can Seedance 2.5 generate a 30-second video?

Yes. ByteDance advertises 30-second continuous output, and the reAPI route accepts 4–30 seconds plus automatic duration.^[1]^[8]

Does Seedance 2.5 support native 4K?

Not on the current reAPI route. Seedance 2.5 accepts 480p and 720p; Seedance 2.0 remains the option for documented 1080p or 4K output.^[2]^[8]

Does it accept 30 images, 10 videos, and 10 audio clips?

Yes on reAPI. The live request schema permits up to 30 images, 10 videos, and 10 audio clips.^[8]

Does Seedance 2.5 integrate with Maya or Blender?

No official plug-in or integration guide was found. References derived from 3D workflows may be supported, but that is not the same as native Maya or Blender integration.

Is Seedance 2.5 better than MiniMax H3?

There is no reproducible benchmark cited in this article that makes one universally better. Seedance 2.5 offers longer single-pass output and explicit edit/extend task types; H3 offers 2K output, mixed references, and native stereo audio.^[5]^[6]^[8]

The honest launch reading

Seedance 2.5 looks like a meaningful workflow upgrade, especially for longer scenes, assigned multimodal references, timestamped direction, and edits to existing footage. Those are the features worth watching.

The accurate headline today is not “30-second native 4K is live.” It is: Seedance 2.5 is live on reAPI for 480p, 720p and 1080p generation, with a 30-second ceiling, documented multimodal limits, and edit/extend workflows. Use doubao-seedance-2.5-face, price source-video jobs with the full billing formula, and keep Seedance 2.0 for higher-resolution output.

References

ModelArk (Volcano Engine). Doubao Seedance 2.5 — 30-second narrative, expanded multimodal references, second-level control, editing, and multilingual demonstrations. Retrieved August 2, 2026 from ark.volcengine.com/promotion?modelName=seedance-2-5
Volcano Engine. Model-list snapshot — Seedance 2.0 duration and resolution capabilities. Retrieved August 2, 2026 from volcengine.com/docs/82379/1330310
Seed Team. Seedance 2.0 Technical Report — multimodal inputs and 4–15-second generation. Published April 20, 2026. arxiv.org/abs/2604.14148
MiniMax. MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities. Published July 31, 2026. minimax.io/blog/minimax-h3
MiniMax API. Video Generation Guide — H3 duration, resolution, native audio, and reference limits. Retrieved August 2, 2026. platform.minimax.io/docs/guides/video-generation
reAPI. Seedance 2.5 — live rates and failed-task refunds. Retrieved August 17, 2026 from reapi.ai/models/seedance-2-5
reAPI. doubao-seedance-2.5-face — public model ID, request limits, resolutions, task types, and source-video billing. Retrieved August 17, 2026 from reapi.ai/docs/seedance-2-5