ByteDance · Announced June 23, 2026 · FORCE Conference
Seedance 2.5
One take. Thirty seconds. Sound included.
Seedance 2.5 is ByteDance's newest video model, and it is live on the Atlas Cloud API today. One native pass of up to 30 seconds with scene changes and tempo shifts, up to 50 multimodal references, and audio generated together with the picture — pay per second, no waitlist.
Seedance 2.5 är utvecklad av ByteDance. Atlas Cloud (som drivs av Atlas Cloud AI LLC) tillhandahåller åtkomst till modellen men äger den inte. Alla varumärken tillhör sina respektive ägare.
Available Now
Three endpoints, one model, one API key
Every Seedance 2.5 variant shares the same generation core, prompt understanding, and audio pipeline. Pick the input you already have — a prompt, a frame, or a reference set — and call it with the same key.
Billed per second of generated video, usage-based, with no minimum commitment. The live rate for the options you pick is shown on each model page before you run anything.
Need native 4K output today, or a lower rate per second? Native 4K is on the way for 2.5; until it lands, the Seedance 2.0 family renders up to native 4K, and its Fast and Mini variants run at lower per-second rates.
Compare the Seedance familyThe Headline Feature
Thirty seconds without a single cut
Most video models top out at a few seconds per pass, and longer stories get stitched together in post. Seedance 2.5 generates the full 30-second arc in one native pass — camera moves, scene changes, and tempo shifts included — while character appearance, lighting, and motion style hold from first frame to last.
Establish
The opening shot sets scene, light, and mood from your prompt and references.
Scene change
The model transitions locations mid-take while characters and grade stay locked.
Tempo shift
Pacing accelerates or breathes — motion and score shift together, in sync.
Payoff
The take resolves as one continuous shot. No seams to hide, nothing to stitch.
What's New in 2.5
Built like a production pipeline, not a clip generator
What Seedance 2.5 adds over the 2.0 family, exactly as it runs on the Atlas Cloud API.
50 multimodal references
Attach up to 30 images, 10 videos, and 10 audio tracks to a single request — 30 seconds of combined audio and video reference — and lock characters, products, locations, and style across the whole take. Seedance 2.0 accepts 9 images and 3 clips each.
Audio-only referencing
New in 2.5: a single BGM, voice, or sound-effect track can be the only reference you supply. The model reads it for pacing, beat matching, and lip sync — no image or video input required.
Sound and picture, generated together
Audio is produced with the visuals rather than layered on after. Dialogue lands on lips, foley lands on frames, and the score follows the cut across the full 30 seconds.
Editing and extension
Change one local element of an existing clip — a product, a costume, the weather — while the rest of the frame stays intact, or extend footage you already have. Both run through the reference-to-video endpoint.
Speech in 10+ languages
Native speech generation across more than ten languages, with stronger instruction following for dialogue and delivery — enough for localized cuts from a single pipeline.
Pipeline-ready controls
Duration from 4 to 30 seconds or auto, 480p and 720p output, six aspect ratios plus adaptive, MP4 or MOV containers, and toggles for audio, watermark, and last-frame return.
Integrate It Today
Two calls from prompt to video
Submit a generation, poll the prediction, read the output URL. Every Atlas Cloud video model takes the same request shape, so moving from Seedance 2.0 to 2.5 is a one-line change.
# 1. Submit the generation
curl -X POST "https://api.atlascloud.ai/api/v1/model/generateVideo" \
-H "Authorization: Bearer $ATLASCLOUD_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "bytedance/seedance-2.5/text-to-video",
"prompt": "A golden retriever running on a sunny beach, cinematic lighting",
"duration": 10,
"resolution": "720p",
"ratio": "16:9",
"generate_audio": true
}'
# 2. Poll until it finishes — outputs[0] is the video URL
curl "https://api.atlascloud.ai/api/v1/model/prediction/PREDICTION_ID" \
-H "Authorization: Bearer $ATLASCLOUD_API_KEY"Swap the model id for bytedance/seedance-2.5/image-to-video or /reference-to-video to use the other endpoints — the rest of the call is identical.
2.0 → 2.5
The spec sheet, side by side
Seedance 2.5 against Seedance 2.0, as both run on Atlas Cloud today.
Seedance 2.5 is newer, not a straight replacement: it doubles clip length, reference capacity, and reference budget, while its resolution ladder opens at 480p and 720p with native 4K to follow — so 4K work belongs on Seedance 2.0 for now. Independent third-party benchmarks for 2.5 have not been published yet. For context, Seedance 2.0 ranks #1 for text-to-video with audio on the Artificial Analysis Video Arena (Elo 1218, June 2026).
Sources: Atlas Cloud model API · ByteDance / Volcano Engine announcement · Artificial Analysis Video Arena
Why Atlas Cloud
The platform 2.5 runs on
Atlas Cloud is where teams run ByteDance video models in production — and where Seedance 2.5 went live the day it opened.
LLMs, image, video, audio, and 3D — a single key and a single invoice.
New models go live the moment they launch. Seedance 2.0 did, and so did 2.5.
Enterprise SLA, SOC 2 compliance, and dedicated infrastructure options.
Switch by changing a base URL — no SDK overhaul to adopt a new model.
Running Seedance 2.5 at scale?
Seedance 2.5 is self-serve — create an account, add a key, and start generating. If you are planning sustained volume, need rate-limit headroom for a launch, or are moving a pipeline over from another provider, our solutions team will size it with you.
- 01Volume pricing and committed-throughput quotes
- 02Dedicated capacity and rate-limit headroom for launches
- 03Migration review from Seedance 2.0 or another video API
- 04A direct line to a solutions engineer, not a ticket queue
You do not need to talk to anyone to start — Seedance 2.5 is available self-serve on the Atlas Cloud API right now. This form is for volume, capacity, and migration conversations.
FAQ
Seedance 2.5, answered
Seedance 2.5 is ByteDance's latest multimodal video generation model, unveiled at the Volcano Engine FORCE conference on June 23, 2026. It generates up to 30 seconds of video in a single native pass, conditions on up to 50 multimodal reference assets, generates synchronized audio with the picture, supports consistency-preserving editing and extension of existing footage, and produces native speech in more than ten languages.
Your next take is thirty seconds long.
Seedance 2.5 is live on Atlas Cloud. One API key, per-second billing, no waitlist.
