Seedance 2.5 Complete Guide: 30-Second AI Video With Native Audio (2026)
Seedance 2.5 generates up to 30 seconds in a single pass with native audio and up to 9 reference images. Full guide to settings, credit costs, prompting, and when to use it.
Seedance 2.5 is ByteDance's latest video model, and it moves two limits that used to define AI video: clip length and audio. A single generation now runs up to 30 seconds, and the soundtrack is produced in the same pass as the picture rather than bolted on afterwards.
This guide covers what changed from Seedance 2, every setting you can control, what each combination actually costs, and how to write prompts that hold together across a 30-second shot.
What Changed From Seedance 2
| Capability | Seedance 2 | Seedance 2.5 |
|---|---|---|
| Max duration | 15s | 30s |
| Native audio | Yes | Yes, included at the same credit cost |
| Reference images | 9 | 9 |
| Start / end frames | 2 | 2 |
| Resolutions | 480p, 720p | 480p, 720p |
The headline is the jump to 30 seconds in one pass. That distinction matters more than the number itself. Stitching three 10-second clips together means three chances for the character's face, the lighting, or the wardrobe to drift. One 30-second generation keeps a single continuous take, which is what makes real camera movement and uninterrupted shot language possible.
Three Ways To Drive It
Seedance 2.5 accepts three different kinds of input, and picking the right one matters more than prompt wording.
- Text to Video — Describe the shot from scratch. Best when you have no existing assets and want full freedom over framing.
- Image to Video — Supply a start frame, and optionally an end frame. With both, the model interpolates a deliberate path between two compositions instead of improvising. This is the most reliable way to control where a shot begins and ends.
- Reference to Video — Supply up to 9 reference images. These are not frames; they teach the model what a character, product, or setting looks like so it stays consistent. This is the mode to use when the same person or product has to appear recognizably throughout.
Resolutions, Durations, and Credit Costs
Seedance 2.5 offers 480p and 720p, at four durations. Costs scale with both, and audio is included in every price below at no surcharge.
| Duration | 480p | 720p |
|---|---|---|
| 5 seconds | 80 credits | 180 credits |
| 10 seconds | 160 credits | 355 credits |
| 15 seconds | 235 credits | 530 credits |
| 30 seconds | 470 credits | 1,055 credits |
A practical workflow follows from that table: iterate at 480p / 5s until the prompt is doing what you want, then re-run the winning prompt at 720p and your target length. Testing a concept at 480p/5s costs 80 credits against 1,055 for a 720p 30-second render, so getting the prompt right first is worth roughly thirteen cheap attempts.
Native Audio Is On By Default
Seedance 2.5 generates its soundtrack alongside the video in the same pass, and it is enabled by default. Because there is no separate audio tier in the pricing above, you pay the same whether you keep the audio or not. If you plan to replace the sound in an editor anyway, there is no cost saving in turning it off, so there is little reason to.
What this means in practice is that ambient sound, room tone, and motion-matched effects arrive already synchronized to the picture. If you need a specific voice or a scripted line instead, generate the video here and then run it through AI Lip Sync or AI Voiceover.
Writing Prompts For a 30-Second Shot
A prompt that works at 5 seconds often falls apart at 30. Short clips only need one idea; long clips need a sequence. Structure the prompt as a timeline instead of a description.
- Break the shot into beats. Say what happens in the first third, the middle, and the end. "0-10s: she walks toward the window. 10-20s: she pauses and looks out. 20-30s: camera pushes past her shoulder to the street below."
- Give the camera one job. A single continuous move — a slow push in, an orbit, a crane down — reads as intentional. Multiple conflicting instructions produce drift.
- State what stays constant. Naming the wardrobe, hair, and lighting explicitly gives the model something to hold onto across 30 seconds.
- Describe the sound. Since audio is generated with the picture, mentioning the sonic environment ("rain against glass, distant traffic, no music") shapes the result.
Where Seedance 2.5 Is Not The Right Choice
Being straightforward about the limits saves you credits.
- You need 1080p. Seedance 2.5 tops out at 720p. For full HD today, use Kling 3.0 (1080p at 5s or 10s, with optional audio).
- You are animating a single photo cheaply. Hailuo 2.3 starts at 25 credits for a 768P 6-second clip, a fraction of Seedance 2.5's entry price.
- You only need a quick draft. Seedance 2 Mini covers 5s and 10s from 45 credits and shares the same 9-reference-image support.
- You want to restyle footage you already have. That is a video-to-video job — use Kling 3.0 Motion Control.
Choosing Between The Seedance Models
| Seedance 2 Mini | Seedance 2.5 | |
|---|---|---|
| Durations | 5s, 10s | 5s, 10s, 15s, 30s |
| Entry cost | 45 credits (480p/5s) | 80 credits (480p/5s) |
| Reference images | 9 | 9 |
| Best for | Volume drafts, social cutdowns | Continuous long takes, dialogue scenes |
Get Started
Seedance 2.5 is live now across Text to Video, Image to Video, and Reference to Video. New accounts include free credits, which is enough to run several 480p tests before committing to a long render. See pricing for credit packs.
For a wider view of how 30-second generation compares across vendors, read The 30-Second Era: Seedance 2.5 vs Wan 3.0 vs FLUX 3 Video.