Seedance 2.5
ByteDance's video model: up to 30 seconds of video with sound in one go, guided by dozens of reference images, clips and sounds.
Quality
Modality
video
Resolution
1080p
Access
closed
Fabian's Take
"Seedance 2.5 seems to be dominating video at the moment. Google's Omni is quite decent, but this seems to be the one to beat right now."
Seedance 2.5 is ByteDance’s video model, launched on 31 July 2026. It makes up to 30 seconds of video in one pass, twice what Seedance 2.0 managed, with sound generated alongside the picture and kept in sync.
Built for consistency
What sets it apart is how much you can give it to work from: up to 30 reference images, 10 video clips and 10 audio clips in a single request. That’s how you keep a character’s face, a product or a voice the same across a clip, which is where video models usually fall apart. You can extend a clip over several rounds, and ByteDance says characters, places and story stay consistent, up to several minutes of video. You can also edit at a specific point in time, swap a background in green-screen style or change the camera angle.
What it costs
Through ByteDance’s BytePlus API, Seedance 2.5 is billed per token, and the tokens depend on length, resolution and frame rate: $10.70 per million tokens, or $6.40 when you feed it video. BytePlus’s own examples put a 5-second clip at about $0.55 in 480p, $1.24 in 720p and $3.06 in 1080p, more when you add reference video. On platforms you pay in their credits instead: on Higgsfield it starts with the Plus plan.
Where you’ll find it
ByteDance’s own apps, Jimeng and Doubao, are aimed at China. Elsewhere it’s in CapCut’s Dreamina, in creative platforms like Higgsfield, and in the BytePlus API for developers. Through Higgsfield, an AI agent like Claude can use it for you as well.
Where it stands
Right now it seems to be the video model to beat. Gemini Omni is quite decent and the easier choice inside Google’s apps.
The Verdict
Best for: Longer, consistent clips with sound: ads, explainers and short scenes where characters and places have to stay the same from shot to shot.
Pros
- Up to 30 seconds per clip, twice Seedance 2.0's 15
- Sound and picture in one pass, kept in sync; reference voices stay the same
- Takes up to 30 images, 10 video clips and 10 audio clips as references
- Extends a clip over several rounds while keeping characters and places consistent, up to several minutes
- Targeted edits at a timestamp, plus green-screen and camera-angle editing
Cons
- Expensive at 1080p, and the token-based price is hard to predict
- On Higgsfield it starts with the Plus plan
- ByteDance's own apps (Jimeng, Doubao) are aimed at China; elsewhere you use it through Dreamina, platforms like Higgsfield or the BytePlus API
Specs
- Pricing Billed per token: $10.70/M without video input, $6.40/M with video input · a 5-second clip from about $0.55 (480p), $1.24 (720p) or $3.06 (1080p) through BytePlus
- Cost Tier premium
- ⚡Speed Tier moderate
- License Proprietary