
Seedance 2.5AI video generator with sound
Up to 30 seconds in one continuous take, with dialogue, effects and music made in the same pass. Direct it with up to 50 image, video and audio references.
No watermark. You see the price before you generate.

What it does
Seedance 2.5 features: one model for the whole shoot

A whole scene in one take, sound included
Stitch short AI clips together and faces, light and props can drift at every seam. Seedance 2.5 holds one take for up to 30 seconds, so the same rider crosses the city from morning to sunset. The bike bells, the traffic and the score are generated with the picture, so they land on the action instead of being laid on afterwards.

Block it in 480p, finish it in 1080p
Turn on Draft to test the timing, the framing and the camera move at 480p while you save credits. When a take works, finish that one at 1080p. The final render is charged separately, so full quality goes only to the shots you keep.
- 480p Draft
- 1080p final

Multimodal references: cast it, dress it, then roll
Give Seedance 2.5 a face, a coat, a pair of shoes and a street, and it builds the shot around them. One generation takes up to 30 images, 10 videos and 10 audio clips: stills to lock a character or a product, videos to copy a camera move, audio to set a voice or a beat. Say in the prompt what each file is for.
Start generating



Built for
Seedance 2.5 for ads, stories and social video
- Developer
- ByteDance Seed
- Length
- 4 to 30 seconds per generation, or Auto
- Resolution
- 480p or 720p. Draft at 480p, then finish the take at 1080p
- Start from
- A text prompt, a first and last frame, or references
- References
- Up to 30 images, 10 videos and 10 audio clips. Clips run 2 to 30 seconds each
- Sound
- Dialogue, effects and music generated with the picture. On by default
- Aspect ratio
- Adaptive, 1:1, 3:4, 4:3, 9:16, 16:9 or 21:9
- Where to use it
- Paper Gen and the canvas in VibePaper
How to generate
Make a Seedance 2.5 video in three steps
Real Paper Gen recording1Add a reference
Optional: a first frame, a face, a product or a place to set the look and the cast.
Real Paper Gen recording2Write the prompt
Describe the subject, the action, the camera move, the light and the sound you want.
Real Paper Gen recording3Generate and download
Check the price on the button, generate, and download the take without a watermark.
Recorded by the VibePaper team in Paper Gen on 12 November 2026. The prompt, the 4-second setting, the credit price on the button and the finished clip all come from one real run, not a mock-up.
Made with Seedance 2.5
Start generatingSeedance 2.5 questions: length, sound, references and price
What is Seedance 2.5?
Seedance 2.5 is a video model from ByteDance Seed. It generates up to 30 seconds of video with synchronized sound in one pass, from a text prompt, a first and last frame, or image, video and audio references. In VibePaper it runs in Paper Gen and on the canvas.
What is new compared with Seedance 2.0?
Longer takes and more reference material. Seedance 2.5 runs up to 30 seconds against 15 for Seedance 2.0, and takes up to 30 images, 10 videos and 10 audio clips against 9, 3 and 3.
How long can a video be, and at what resolution?
Each generation runs 4 to 30 seconds at 480p or 720p. To save credits while you test, generate a 480p Draft first, then finish the one you like at 1080p. The final render is charged separately.
How much does a Seedance 2.5 video cost?
The price depends on length and resolution, and the Generate button shows it before you commit. In our own run in Paper Gen on 12 November 2026, a 4-second 720p video with one image reference showed 60.94 credits; the same prompt left on Auto length showed about 259, so set the length yourself when you are testing.
How long does Seedance 2.5 take to generate a video?
It depends on the queue and the length. The 4-second 720p video we recorded for the steps above took about 10 minutes to finish. A 480p Draft is the quicker way to check a shot before you spend on the final.
Does Seedance 2.5 generate sound?
Yes. Dialogue, ambient sound, effects and music are generated with the picture, so they follow the action. Sound is on by default and you can turn it off before you generate.
How many references can I use?
Up to 30 images, 10 videos and 10 audio clips in one generation. Videos and audio clips run 2 to 30 seconds each. Tell the prompt what each file is for, such as the face, the coat or the street.
Can it keep the same character across shots?
Yes, in one generation. Describe the character the same way in every shot, or add a reference image for the face and outfit. Keep the cast small when several faces have to stay consistent.
Why was my prompt refused?
Prompts that name a known film, song, artwork or real person can be refused. Describe original characters and places in your own words instead.
Can I use the videos commercially?
Yes. Videos download without a watermark, and you can use them in ads, client work and social channels. You are responsible for the images, footage and voices you upload as references.
Guides and Model Briefs from Our Blog
Practical guides and plain-language briefs on the models you can use here: what each one is good at, how to prompt it, and how to get a better result on the canvas.











