videoPowered by ByteDance

Seedance 2.5

ByteDance’s video model. Give it a prompt or an image and it makes up to 30 seconds of video in one shot, with sound that comes out already in sync.

Shown at the 2026 Volcano Engine FORCE conference

30s
Native single shot
50
Reference inputs
2K
Cinematic output
Synced
Audio and video

What’s new

30 seconds in one take

Seedance 2.5 renders a full 30 seconds as a single continuous shot, about double the last version. The camera move and the action hold steady from the first frame to the last, so you stop stitching short clips together to fake a long scene.

Up to 50 references

Feed it as many as 50 reference files at once: images, video, audio, storyboards, character sheets. It keeps a face, a product, or a style intact instead of drifting partway through the clip.

Edit one part, keep the rest

Change a single object, swap an outfit, or relight one area without re-rendering the whole clip. The composition, camera path, and background stay exactly where they were, so a fix is a tweak instead of a fresh roll.

Built for real production

2K output

It renders at a clean cinematic 2K, so footage holds up on a big screen instead of falling apart under close inspection. When you need more, the upgraded 2.0 line now outputs native 4K.

Steadier motion

Motion is one of the clearest wins in Seedance 2.5. Faces, hands, and backgrounds show far less flicker and warping than older models, so people stay people and lines stay straight.

Audio that fits

Seedance 2.5 renders picture and sound in one pass, so you never bolt audio on after the fact. Dialogue lands with close lip-sync, and effects and music sit under the action.

Camera moves

Ask for a push-in, a dolly, a pan, an orbit, or a slow-motion beat and Seedance 2.5 follows it. The moves read like real camera work, which makes a 30-second shot feel shot rather than generated.

Consistent characters

A face, a product, and a set of brand colors stay the same from the first frame to the last. Seedance 2.5 leans on your reference files, so a character will not quietly change outfit or hairstyle partway through.

Three ways in

Start from a line of text, animate a still image, or guide the model with your own reference files. Whichever way you go in, the look and motion quality stay consistent on the way out.

More it can do

AI Dance

Drop in one photo of a person, pet, or avatar and it animates a full dance from a built-in routine library, without limbs bending the wrong way.

Better color

Color reads about 15% more accurately than the last version, and skin tones stop sliding toward the odd casts that gave older AI video away.

50 shot templates

Pick from 50 preset camera setups. Type one line of direction and it matches the move, from a slow push-in to an orbit.

Sound and picture together

A two-branch design renders audio and video at the same time, so on-screen action sits on the soundtrack without a manual sync pass in post.

What people make with it

Social clips for TikTok, Reels, and ShortsProduct and e-commerce adsShort dramas and story scenesPre-viz for film and TVBranded videos with a consistent characterMusic videos and dance clipsVirtual influencers and mascots

FAQ

1

What is Seedance 2.5?

It is ByteDance’s video model. Give it text, an image, or reference files and it makes up to 30 seconds of video with sound, steady characters, and real camera moves.

2

Who made Seedance 2.5?

ByteDance built it, and it runs on the Volcano Engine platform. The model was shown at the 2026 Volcano Engine FORCE conference.

3

How long can a clip be?

Up to 30 seconds in a single continuous shot, roughly twice the previous version, with no visible cuts.

4

Does it generate audio?

Yes. It renders sound and video at the same time, so effects, dialogue, and music line up with the picture, lip-sync included.

5

How is it different from Seedance 2.0?

It doubles the single-shot length to 30 seconds, takes up to 50 reference files, edits one region at a time, and reads color more accurately. The 2.0 line picked up native 4K in the same update.

6

Can I edit a clip without redoing it?

Yes. Replace an object, change an outfit, or relight one area while the rest of the shot stays put.

7

What resolution does it output?

Up to 2K, with steadier skin tones. The companion 2.0 line handles native 4K.

8

How many references can it take?

Up to 50 in one go: images, video, audio, storyboards, and character sheets.

Make your first Seedance 2.5 clip

Type a prompt or drop in an image. You get cinematic video with sound, straight from the browser.

Try Seedance 2.5