Skip to main content
Seedance 2.5 is ByteDance’s joint audio-and-video model in Tess. In one pass it produces a clip of up to 30 seconds, with speech, music, and effects timed to the picture, at 480p or 720p.
Seedance 2.5 is a video model, not a conversational LLM. You use it in Chat, through the Video tool, and in the Video Generator.

​
What 2.5 delivers in Tess

With a first frame or with references, the aspect ratio used is adaptive: the tool follows the media.

​
How to use it in Chat

1

Open the video tool

In Chat, open Tools and go to Video.
2

Describe the clip

Include the subject, action, setting, camera, duration, and aspect ratio. For speech, put the spoken line in quotes.
3

Attach media, if you have any

Use a first frame or references, and say in the prompt what each file is for: product, camera, dialogue, or score.
4

Revise in the same conversation

Ask for the change and state what must stay the same.

​
How to use it in the Video Generator

In the left sidebar, open Agent Studio and then Video Generator. Write the prompt and set duration, aspect ratio, and resolution before you generate. See Video Generator.
In the interface, the tool is named Seedance PRO. Open that tool and choose Seedance 2.5 from the available models.
Include duration, aspect ratio, and resolution in the request. For example: “Generate 12 seconds in 9:16, 720p, with audio.”

​
References

One generation accepts at most 30 images, 10 videos, and 10 audio files (WAV or MP3). Reference videos and reference audio each add up to at most 30 seconds. Reference audio needs at least one reference image or reference video. In the prompt, tag each file as [Image1], [Video1], [Audio1]. First frame, and first plus last frame, are a separate path. The end image only works when the first frame is set. That path and the references do not go in the same generation. If the first and last frames have different aspect ratios, the last frame is cropped to the first.
A direct upload of reference images or videos that contain a real person’s face is not accepted.

​
First frame

​
Product reference

​
Where Seedance 2.5 is most useful

  • Reels and ads with a beginning, middle, and end in the same clip
  • Scenes with several subjects, each with a role in the prompt
  • A product or a style locked to a reference image
  • A clip with speech, effects, and room tone in the same file

​
Best practices

  • Do not mix a first frame with reference images, videos, or audio.
  • Keep reference videos and reference audio each within a combined 30 seconds.
  • State what changes and what stays.
  • Review hands, in-frame text, brands, and continuity before you publish.
Credit use depends on duration, resolution, and whether a reference video is attached. The current price is on Models and Costs. See also: Video Generation · Video Generator · Seedance 2.5.