ByteDance · BytePlusVIDEOAnnounced · not bookable

Seedance 2.5 is not on VisionX yet.

Announced: a native 30-second single take, up to 50 references, 4K at 10-bit.

01

What Seedance 2.5 is, and is not

THE MODEL

Seedance 2.5 is the next Seedance generation, announced by ByteDance in June 2026. It is built around a different shape of output from 2.0: one continuous 30-second take rather than several shots inside a shorter clip.

It is not on the VisionX roster yet. You cannot run it, book against it, or price it today, and nothing on this page has been verified on our own infrastructure — every figure here is the vendor’s own claim. This page exists so you can plan around it. Seedance 2.0 is what runs in the composer right now.

A 30-second single take

One continuous shot with no stitching and no scene-cut splicing, against the multi-shot-inside-a-shorter-clip shape that Seedance 2.0 produces today.

Up to 50 reference inputs

Images, audio, video and style references combined in one generation, against the twelve Seedance 2.0 accepts.

Native 4K at 10-bit

Generated at 4K with 10-bit colour depth. The bit depth is the part that matters, because it is what gives a grade headroom to work in.

02

Seedance 2.5 — announced, not bookable

ON THE HORIZON

This is what ByteDance announced. VisionX has not benchmarked any of it, none of it is in the composer, and none of it has a VX price.

Announced · not on the roster yet

30-second native single shot

One continuous take with no stitching and no scene-cut splicing — a different shape of output from 2.0, which puts several shots inside a shorter clip.

Up to 50 reference inputs

Images, audio, video and style references combined, against the twelve that Seedance 2.0 accepts today.

Native 4K at 10-bit

Generated at 4K with 10-bit colour depth, which is the part that matters for grading headroom rather than for pixel count.

Joint audio-video in one latent space

Sound and motion co-processed rather than aligned afterwards — the same principle as 2.0, pushed further.

Roughly 20% better prompt adherence

ByteDance's own reported figure against Seedance 2.0, which if it holds means fewer re-rolls to a usable take.

Reframe from one master

Recompose a single 4K master into vertical and square crops without paying for a second generation.

Announced by ByteDance at the Volcano Engine FORCE conference, 23 June 2026. Every figure above is the vendor’s own keynote claim and is pending independent verification at public launch — VisionX has not benchmarked this model, and these numbers may not survive contact with a real render.

Seedance 2.0 runs on Standard and Cinema today, with native 4K on Cinema and multi-shot output inside a single generation. When 2.5 reaches general availability it joins the roster the same way every other engine does: same Cast, same wallet, quoted per shot before you run it. Because identity lives in your Cast rather than in any one model, moving a campaign onto it later will not mean rebuilding the campaign.

03

What Seedance 2.5 would be for

USE CASES

What a 30-second single take would be for, if the announced specs hold. These are use cases for the shape of the output, not benchmarks.

Unbroken social hero films

One continuous take for vertical placements, where a cut is the thing that breaks the illusion.

Product ads from a single frame

A packshot carried through a full-length spot without the mid-way identity drift that stitching invites.

Branded story films

A complete narrative beat in one unbroken shot rather than a sequence assembled afterwards.

Talking-head and presenter content

A portrait plus an audio track, held for long enough to carry a real script rather than a single line.

Beat-matched music visuals

Motion and cuts aligned to a track across a full-length run instead of across a few seconds.

Multi-character scenes

A deep reference budget is what would keep cast, wardrobe and props consistent across a long take.

04

Seedance 2.5 against the alternatives

COMPARISON

How the announced Seedance 2.5 figures line up against two engines you can actually run on VisionX today. Read the first column as a vendor claim, not a measurement.

Seedance 2.5 capability comparison against Kling 3.0 and Veo 3.1
CapabilitySeedance 2.5 (announced)Kling 3.0Veo 3.1
Max single-shot duration30s, claimed15s8s
Reference inputsUp to 50, claimedText, imageText, image
Native audioJoint audio-video, claimedSeparate sync stepSame pass as the picture
Max resolution4K at 10-bit, claimed4KUp to 4K
Runnable on VisionX todayNoYesYes

Seedance 2.5 figures are ByteDance’s June 2026 announcement and are pending independent verification at public launch. Kling and Veo rows describe those vendors’ published capabilities as of July 2026 and are the two engines here you can route a shot to right now.

FAQ

Seedance 2.5, answered

Can I use Seedance 2.5 on VisionX today?
No. Seedance 2.5 is announced but not on the VisionX roster, so it is not in the composer, has no VX price, and cannot be booked or scheduled against. Seedance 2.0 is the current generation and runs today across four tiers, up to native 4K on Cinema.
Is the Seedance 2.5 API available yet?
Partly. ByteDance announced Seedance 2.5 at the Volcano Engine FORCE conference on 23 June 2026, and API access began rolling out through Volcano Engine in the weeks after. There is still no confirmed international availability date, and Seedance 2.5 does not appear in the BytePlus ModelArk model catalogue — which is the route VisionX runs the core Seedance tiers through. That is the specific reason it is not on our roster: we cannot call a model our provider does not list.
When will Seedance 2.5 be on VisionX?
When BytePlus lists it and we can call it. We are not going to announce a date we do not control. What we can commit to is that adding it will not disturb anything you have already built: identity lives in your Cast, so a new engine is a routing choice rather than a migration.
How is Seedance 2.5 different from Seedance 2.0?
The headline difference announced is shape of output: 2.5 is built around one continuous 30-second take, where 2.0 puts several shots inside a shorter clip. ByteDance has also announced a much deeper reference budget, 4K at 10-bit, and roughly 20% better prompt adherence. All of those are vendor claims that we have not verified.
Are these Seedance 2.5 specifications confirmed?
They are ByteDance’s own stated figures, not measurements. Everything here comes from the June 2026 Volcano Engine keynote and the vendor material that followed it, and no benchmark has been published for the prompt-adherence claim in particular. VisionX has not run this model, so we have verified none of it ourselves. We publish the numbers attributed and caveated rather than not at all, because planning around an announced model is reasonable — booking against one is not.
What should I run instead right now?
Seedance 2.0 on the Standard or Cinema tier. Cinema renders native 4K directly with no upscale pass, both tiers accept up to twelve references and generate synchronised audio in the same pass as the picture, and both fuse finished clips into one continuous video when a shot needs to run longer.
Will moving to Seedance 2.5 later mean redoing my campaign?
No. Identity on VisionX lives in your Cast, not in any one model, which is the whole point of the roster. When a new engine arrives you route shots to it and the same face, wardrobe and grade resolve — the campaign does not get rebuilt around the model.

Seedance 2.0 runs today.

Seedance 2.5 is not on the roster yet. 20 VX free on signup, no card required — spend it on what you can actually render.