Dialogue that lands on the lips
Quote the line. The mouth moves to match.
BYTEDANCE · SEED VIDEO MODEL · NOW ON CREATIDE AI
Thirty seconds in one take. Fifty references in one pass. Sound born with the shot.
Announced June 23, 2026 · Last updated
Sources:ByteDance Seed (June 2026)BytePlus pricing (Nov 2026)
Definition
ByteDance Seedance 2.5 is an AI video model built for long, controlled takes. It turns a prompt, a start frame or a stack of references into up to 30 seconds of continuous video, with sound, in a single generation.
Announced by ByteDance on June 23, 2026 at Volcano Engine FORCE, Seedance 2.5 doubles the single take from 15 to 30 seconds, takes up to 50 references per request and generates dialogue, effects and ambience with the picture.
You direct a whole scene as one shot, then fix one region without re-rolling. On Creatide AI, the Seedance 2.5 AI video generator sits next to every leading model, with 2× duration (beta) and 8K output.
From ByteDance, the home of TikTok
Seedance 2.5 comes from ByteDance Seed, the AI research team at the company behind TikTok and CapCut. You can feel it in the output: the model thinks in vertical frames, opens on a hook and lands sound on the cut. One take runs up to 30 seconds, long enough for a full post in a single shot.
Need the cover too? Make it with Seedream 5.5, ByteDance’s image model, then bring it to life here.
One continuous take
A 30-second AI video is a full ad spot or a whole scene. Seedance 2.5 renders it in one pass, so light, wardrobe and camera stay continuous.
Write beats with timestamps and the model follows them.
Between 4 and 30 seconds per generation. On Creatide AI, 2× duration (beta) takes the same model to 60 seconds in one take.
Native audio
Seedance 2.5 generates sound with the picture. A door slams on the frame it closes.
Footsteps, engines, rain, score and dialogue, generated together with the video.

Quote the line. The mouth moves to match.
Footsteps and engines follow the action.
Use a voice or track as a reference.
References
Seedance 2.5 references can be faces, products, wardrobe, locations, style frames, a camera move or a voice. Seedance 2.5 keeps every one steady for all 30 seconds.
30 IMAGES · 10 VIDEOS · 10 AUDIO CLIPS · ONE REQUEST
Yes. A few face angles and a wardrobe reference keep the character steady through every scene change.
Region editing
Describe the fix and the model repaints only that region. Every frame you already liked stays as it was.
Turn on 2× duration (beta) and one Seedance 2.5 prompt runs a full minute in a single take. Every take exports at 1080p, 2K, 4K, 6K or 8K: 33 million pixels a frame, enough to crop, reframe and still hold a billboard.
Prompt guide
Break the 30 seconds into beats, each with a time range, a camera move and a sound.
One line per beat, each with a time range.
Handheld, dolly, crane, push-in. The camera is half of the shot.
Quote the dialogue and name the sounds. They are generated with the picture.
Faces and products go in references, not adjectives.
[00–08s] A courier in a yellow raincoat weaves through a night market. Handheld, 35mm, neon on wet stone.
[08–16s] She ducks under an awning. Push-in. Noodle-stall steam, warm tungsten light.
[16–24s] Close-up. She says: “Last delivery tonight.” Rain hiss, distant thunder.
[24–30s] She steps back into the rain. Crane up. Score swells, then cuts.
How to
Pick Seedance 2.5 in the model menu.
Up to 50 references, or one start photo.
Use time ranges, up to 30 seconds or 60 with 2× (beta).
9:16 to 21:9, 1080p to 8K. Fix regions after.
Made for
The formats that get the most out of 30 seconds.
Hook, product and end card in one pass.
A whole scene as one shot.
Driven by your track as an audio reference.
Talking heads in 10+ languages.
Pricing
Same 30 seconds at 1080p, a third of the price of the official Seedance 2.5 API on BytePlus.
Creatide AI · 30s · 1080p
Official BytePlus API · 30s · 1080p
List prices for Seedance 2.5, 16:9, no video input, checked November 2026. Failed generations refund automatically.
Compare
2.0 for short, punchy clips. 2.5 for long takes, heavy references and fixes.
Variants
Same family, two jobs. Flash drafts fast and cheap. The full model finishes the take in 1080p with every reference.
480p or 720p, up to 30 seconds, native audio, 20 references. About a third cheaper.
$0.152/s official API · 1.5× less
Best for drafts, hook tests, social volume
Up to 1080p, 30 seconds, 50 references and region editing. The one you ship.
$0.569/s official API · 3.9× less
Best for ads, client work, long takes
Draft mode: preview a take at 480p, then render the keeper at full resolution. The preview bills at the 480p rate.
Choosing a model
Where Seedance 2.5 shines, and when another model fits better.
Showcase
The Creatide AI difference
It is the same Seedance 2.5 that ByteDance ships. Creatide AI gives it more room: longer takes, bigger exports and a chat that edits for you.
Flip the switch to compare. Official limits from ByteDance, November 2026.
Plans
One take runs up to 60 seconds with 2× duration (beta) and exports up to 8K. Bigger plans add 4K, 8K, more clips and cheaper credits.
Best value for casual creators.
Billed for 12 months. Cancel anytime.
Save $59 compared to monthly
Maximum savings for professionals.
Billed for 12 months. Cancel anytime.
Save $373 compared to monthly
The ultimate suite for elite experts.
Billed for 12 months. Cancel anytime.
Save $673 compared to monthly
Not into subscriptions? Start with a one-time credit pack
FAQ
ByteDance’s AI video model from the Seed family. It generates up to 30 seconds of video with native audio in one pass, takes 50 references and repaints single regions.
ByteDance announced it on June 23, 2026 at Volcano Engine FORCE in Beijing, with general availability from early July 2026.
From 4 to 30 seconds per generation. On Creatide AI, 2× duration (beta) extends a single Seedance 2.5 take to 60 seconds.
1080p, 2K, 4K, 6K or 8K depending on your plan, from 9:16 vertical to 21:9 widescreen.
Yes. Dialogue, effects, ambience and music are generated with the video, and characters speak 10+ languages.
Up to 50 per request: 30 images, 10 video clips and 10 audio clips. You can also drive a generation from audio references alone.
Yes. Add face, wardrobe and product references and identity holds across the whole take.
A faster, cheaper Seedance 2.5 tier from November 2026. It renders 480p or 720p takes up to 30 seconds with native audio and 20 references, at about $0.152 a second at 720p on the official API.
30 seconds at 1080p costs $4.34 on Creatide AI versus $17.06 on the official BytePlus API, about 3× less. Paid plans include commercial rights.
Every video model runs twice as long.
Producer controls for AI music.
Call camera, lens and light with taps.
Export any Seedance 2.5 take up to 8K.
Every leading video model behind one prompt bar.
Ten keyframes and fast action in 30 seconds.
2K video with stereo sound from 12 inputs.
Native 4K with voice references.
Google’s model that you edit by chat.
Turn a 3 to 9 panel storyboard into one video.
Turn decks, PDFs and web pages into video.
Keyframes, references and lip-synced voices.
Open Seedance 2.5, write the beats, press generate.
Seedance 2.5 · 3× cheaper than the official API · Commercial rights on paid plans