What Is Seedance 2.5? Specs, Limits and Real Prices
What is Seedance 2.5: the 30-second ByteDance video model announced 31 July 2026. Full field list, reference media budgets, per-second rates and hard limits.
Seedance 2.5 is the current version of ByteDance's video generation model, announced on 31 July 2026. Asking what is Seedance 2.5 in practice means asking three things: how long a clip it makes, what it will take as input, and what a finished second costs. The answers are 4 to 30 seconds, a wide set of reference media, and a rate that swings by a factor of four across resolutions.
What is Seedance 2.5 and what changed in it
Three differences against Seedance 2.0 survive into the shipped model.
Clip length doubled. One Seedance 2.5 generation runs up to 30 seconds, where 2.0 stops at 15. Multiple shots fit inside that window, so a request behaves like a short scene rather than a single take.
Reference input widened considerably. A request accepts up to 30 reference images, 10 reference clips and 10 audio tracks at once.
Editing and extending arrived. Given a source clip, the model can alter that footage or continue it forward instead of only generating something new beside it.
Native audio, multi-shot scenes and realistic motion carried over from 2.0 without change.
The specification
| Field | Accepted values |
|---|---|
| Duration | 4 to 30 seconds, default 5 |
| Resolution | 480p, 720p, 1080p, default 720p |
| Aspect ratio | 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, adaptive (default) |
| Prompt | up to 20,000 characters |
| Reference images | up to 30 |
| Reference videos | up to 10 clips, 2 to 30 seconds each, 30 seconds combined |
| Reference audio | up to 10 tracks |
| Audio generation | on by default |
| Content filter | on by default |
The prompt is required for text-to-video and optional once reference media is attached. Media is supplied as public HTTP(S) URLs; data URIs are rejected. Setting duration to -1 lets the model choose the length, and the reservation is taken at the 30-second cap.
Sound comes out with the picture
Audio is generated in the same pass as the video, covering speech, sound effects and music, delivered as a mono track. There is no separate voice step and no lip-sync pass to run afterwards.
Dialogue is written into the prompt inside quotation marks. Whatever sits between the quotes is what gets spoken, so the line has to be short enough to be said inside the duration that was requested. Turning audio generation off produces a silent clip.
What a Seedance 2.5 clip costs
Billing is per second of finished video, at a rate set by resolution and by whether the request carried a reference clip.
| Resolution | Generation only | With a reference video |
|---|---|---|
| 480p | $0.118589 | $0.071153 |
| 720p | $0.266824 | $0.160094 |
| 1080p | $0.461856 | $0.275904 |
Ten seconds of 1080p generated from scratch is about $4.62. The full 30 seconds at 1080p is about $13.86. The same 30 seconds at 480p is about $3.56, which is why draft passes belong at the lower tier.
Reference images and audio tracks are free. Reference video is billed for its own length on top of the output, and a request containing reference video is never billed for fewer than five thirds of the output duration, rounded up. A six-second reference clip with a ten-second 1080p output bills 17 seconds rather than 16, coming to about $4.69.
Generation averages two to five minutes, the job is asynchronous, polling for the result is free, and failed jobs are refunded in full.
The limits
Resolution stops at 1080p. Pre-release material described a 4K tier and it never shipped, so any spec table listing 4K for this model is quoting an announcement. No host, plan or parameter produces it.
Thirty seconds is the hard ceiling on a single generation. Longer pieces are chained segments, which is true of every model in this class.
Named real celebrities, third-party intellectual property and illegal content are blocked inside the model, and no host setting removes them. The content_filter field defaults to true; setting it false selects a less restrictive channel and leaves moderation in place. Refusals return error 80006, which does not say which stage fired.
Frequently asked questions
When was Seedance 2.5 released? ByteDance announced it on 31 July 2026. Availability through hosted APIs followed, so the date it became callable depends on the platform.
How long can a Seedance 2.5 video be? Four to thirty seconds in one generation, with a default of five. Several shots can sit inside that window.
Does Seedance 2.5 support 4K? No. The resolution options are 480p, 720p and 1080p. The 4K figure comes from pre-release coverage of the model and was never delivered.
How much does a Seedance 2.5 video cost? At 1080p, about $0.46 per second of output, so roughly $4.62 for ten seconds. At 480p the same ten seconds is about $1.19. Reference video changes the calculation by adding its own seconds to the bill.
The short version
What is Seedance 2.5: ByteDance's 31 July 2026 video model, generating 4 to 30 seconds of video with synchronized mono audio in one pass, from a prompt of up to 20,000 characters plus as many as 30 reference images, 10 clips and 10 audio tracks. It bills by the second, tops out at 1080p with no 4K anywhere, refunds every failure in full, and keeps moderation on regardless of how the content filter is set.
