THE WORLD'S BEST VIDEO MODEL

Seedance 2.5Unlimited. Just $49/month.

ByteDance's new flagship: 30-second scenes, sound built in. Unlimited on Infer for $49/month — the best plan in AI video. Free for 24 hours.

UNLIMITED · $49 / MOOUTPUT · 30s · sound built in
Try it now
One unbroken take: the camera follows a courier through a lantern-lit night market, vendors calling out, rain just ended.Generated by Seedance 2.5 · ByteDance
Live output
Two friends argue quietly across a café table — “You knew, didn’t you?” — rack focus between faces, espresso machine hissing.
A dancer hits every beat of the track on a rooftop at blue hour, city bokeh behind, camera locked off.
Storm waves burst against the lighthouse as the beam sweeps through sheets of rain, foghorn low in the mix.
03 / Use cases

Where Seedance 2.5 shines.

Dialogue scenes

Write the lines in quotes. The model performs them, lip-synced.

EXAMPLETwo friends argue quietly across a café table — “You knew, didn’t you?” — rack focus between faces, espresso machine hissing.

UGC ad creative

Hook, demo, CTA — a complete vertical ad in one 30-second take.

EXAMPLECreator talks straight to camera about her morning routine, handheld selfie framing, kitchen daylight, natural delivery.

Consistent characters

Lock a face across every shot with up to 30 reference images per run.

EXAMPLEThe red-jacketed courier from the reference set weaves her bike through evening traffic, face steady in every frame.

Music-driven clips

Movement lands on the beat, not near it. Feed it up to 10 reference tracks.

EXAMPLEA dancer hits every beat of the track on a rooftop at blue hour, city bokeh behind, camera locked off.

Previz & blocking

Block with white-models and references, then render the real scene.

EXAMPLEA single steadicam follows two detectives arguing down the precinct corridor, phones ringing behind office doors.

One-take camera moves

Follows, orbits, and slow reveals no 5-second model can hold.

EXAMPLEOne unbroken take: the camera follows a courier through a lantern-lit night market, vendors calling out, rain just ended.

06 / Examples

Generated with Seedance 2.5.

A live cross-section of the model's range — portraits, products, typography, illustration, fashion, cinematic. Hover any tile to read its prompt.

One unbroken take: the camera follows a courier through a lantern-lit night market, vendors calling out, rain just ended.
Two friends argue quietly across a café table — “You knew, didn’t you?” — rack focus between faces, espresso machine hissing.
A dancer hits every beat of the track on a rooftop at blue hour, city bokeh behind, camera locked off.
Storm waves burst against the lighthouse as the beam sweeps through sheets of rain, foghorn low in the mix.
The red-jacketed courier from the reference set weaves her bike through evening traffic, face steady in every frame.
A single steadicam follows two detectives arguing down the precinct corridor, phones ringing behind office doors.
The chef drops noodles into the roiling pot — steam, sizzle, and ladle-clang carrying the scene, slow push-in.
A lone wolf crests the snowfield at dawn, breath steaming, wind and distant ravens in the soundtrack.
An astronaut hauls herself along the station's hull, Earth turning below, breathing loud inside the helmet.
Macro orbit around a mechanical watch as the balance wheel ticks, each tick crisp in the audio.
A paper-craft city folds itself upright block by block, paper rustling with each fold, stop-motion rhythm.
Creator talks straight to camera about her morning routine, handheld selfie framing, kitchen daylight, natural delivery.
07 / Benchmarks

By the numbers.

30 seconds — nearly 4× Veo's ceilingSingle shot
50 files: 30 images, 10 videos, 10 audioReference inputs
Generated in the same pass, lip-syncedSound
VS
Seedance 2.0 Pro3× the clip length, 4× the references, and sound generated in the same pass.
MiniMax H3Twice the single shot (30s vs 15s) and 50 reference inputs vs 15.
Veo 3.1 FastNearly 4× the clip ceiling (30s vs 8s) and 50 reference inputs vs 3.
08 / Pricing
$0.13/ second of video

Pay only for successful generations. No idle, no minimums, no per-seat. Volume discounts kick in at 10K req/mo.

VS NativeOne key, 60+ models, one bill. No ByteDance account.
VS Self-hostClosed weights — Infer is how you run it.
09 / FAQ

Things teams ask.

Q.01What's the $49 offer?
Infer Unlimited: $49/month, all the video and image generation you want in the Studio, Seedance 2.5 included. Free for 24 hours, cancel anytime. API usage is metered separately, per second.
Q.02How long can clips be?
Up to 30 seconds in one shot — the longest single generation of any major model — and you can extend a clip twice.
Q.03Does it really generate audio?
Yes. Dialogue, ambience, and effects generate in the same pass, synced to the picture. Put lines in quotes and they're delivered lip-synced.
Q.04What are reference inputs?
Up to 50 files per request: 30 images to lock identity or style, 10 videos for framing and camera language, 10 audio tracks to steer voice or score.
Q.05What input modes are supported?
Text-to-video, image-to-video from a keyframe, and reference-to-video, which combines your prompt with reference files in one request.
Q.06What resolution is the output?
ByteDance's API launches with 480p and 720p tiers; higher-resolution output has been shown in its own apps. Infer's picker always shows exactly what the live endpoint serves.
Q.07What editing controls does it have?
Region-level edits, green-screen editing, and white-model (3D blocking) control for previz — plus a wider range of audio and visual edit requests than any Seedance before it.
Q.08How do I get the best results?
Prompt like a director: subject, action, setting, one camera move, and the light. Dialogue goes in quotes. Block a 30-second scene beat by beat: Shot 1, Shot 2, Shot 3.
Q.09Can I go longer than 30 seconds?
Yes — extend a generation twice. Framing and subjects stay consistent across the joins.
Q.10Can I use the outputs commercially?
Yes. Same commercial terms as Seedance 2.0 on Infer.
10 / Ready to ship?

Ship with Seedance 2.5.

One key. One bill. One SDK shape — across 50+ models. Pay only for what you use.