Try Seedance 2.5 on Automatica

Seedance 2.5 has landed. ByteDance's newest video model is available in the Automatica model picker today — no waitlist, no separate API key, and billed with the same credits you already use for every other model on the platform.

If you've been working with Seedance 2.0, the headline change is simple: your shots got twice as long, and you stopped stitching them together.

What's actually new

One continuous 30-second take

Seedance 2.0 capped out around 15 seconds, which meant anything longer had to be generated in pieces and cut together — and every cut was a place where a face drifted, a jacket changed shade, or the lighting jumped. Seedance 2.5 generates up to 30 seconds in a single pass. One take, one continuous camera, no seams to hide.

For a product demo, an ad spot, or a dialogue beat, that's the difference between a shot you can ship and a shot you have to fix.

A much bigger reference budget

You can now anchor a generation against a substantially larger set of reference material — images, video clips, audio, and style references, combined in one prompt. Each asset can take a role: this image is the character, this one is the product, this clip is the camera move, this audio is the voice.

The practical effect is consistency. Faces, wardrobe, packaging, and colour grade hold across the full 30 seconds instead of degrading toward the end of the clip.

Native audio, generated with the picture

Audio isn't layered on afterwards. Dialogue, ambience, and effects are generated jointly with the video, so lip sync lands and footsteps hit the ground at the right frame. Multi-language voice is supported, which makes localised ad variants far less painful than re-cutting audio per market.

Editing instead of re-rolling

This is the part that changes how you work. Seedance 2.5 supports region-level and timestamp-level editing — you can change one element inside a shot while the rest of the scene, motion, and style stay exactly as they were. Swap a character, replace a product, adjust an action at a specific second.

You're no longer rolling the dice on a full regeneration and hoping the good parts survive.

Extension and camera control

Generated clips can be extended across multiple rounds to build longer sequences, and camera behaviour — angle, movement, perspective — responds to plain natural-language direction rather than memorised prompt jargon.

Specs at a glance

One honest caveat: resolution moved down, not up. Seedance 2.5 currently tops out at 720p. If you need 1080p or higher for a client deliverable, Seedance 2.0 is still the right call — it stays available on Automatica and isn't going anywhere.

When to use which model

Reach for Seedance 2.5 when:

• The shot needs to run longer than 15 seconds without a cut

• You're locking a character or product across a full scene

• You want to revise one element without regenerating everything

• Dialogue and lip sync matter

Stay on Seedance 2.0 when:

• You need maximum resolution

• The shot is short and quality-per-frame is the priority

Look at Gemini Omni or Veo instead when:

• You want conversational, turn-by-turn refinement

• You're working inside a broader multimodal reasoning workflow

Getting the best out of it

Give every reference a job. More references don't mean better output. Nine well-labelled assets with clear roles beat thirty ambiguous ones — the model needs to know which image is the character and which is the environment.

Write for 30 seconds, not for 5. A prompt built for a short clip leaves the model to invent 25 seconds of filler. Describe the arc: what happens at the start, what shifts in the middle, where it lands.

Direct the camera in plain language. "Slow push in, then hold" works. You don't need lens specifications unless you actually want them.

Edit rather than re-roll. When a generation is 90% right, use the editing routes on the part that isn't. It's cheaper and it protects the parts you liked.

How to start

1. Open the Automatica model picker and select Seedance 2.5

2. Choose your route — text-to-video, image-to-video, or reference-to-video

3. Upload your references and assign each one a role in the prompt

4. Set duration and resolution

5. Generate, then use the editing tools to refine specific regions or moments

Seedance 2.5 is available on all Automatica plans. If you're on Basic and want to run longer generations at volume, Pro and Max give you more monthly credits and earlier access to new models as they land.

Try Seedance 2.5 →

FAQ

What is Seedance 2.5? It's ByteDance's latest multimodal video generation model, announced at the Volcano Engine FORCE conference in June 2026 and released on 31 July 2026, with developer API access following in early August.

How long can Seedance 2.5 videos be? Up to 30 seconds in a single continuous generation, with multi-round extension available for building longer sequences.

Does Seedance 2.5 generate audio? Yes. Audio is generated jointly with the video, including synchronised dialogue and lip sync across multiple languages.

Does it support 4K? No. Despite early reporting, the shipped model supports 480p and 720p. Seedance 2.0 remains the higher-resolution option.

Is Seedance 2.0 still available on Automatica? Yes. Both models stay in the picker, and 2.0 is still the better choice when you need 1080p.

How much does it cost? Seedance 2.5 is billed with Automatica generation credits, priced per second of output. Longer clips consume more credits.