How to use Seedance 2 and edit the clips
Last updated 2026-10-20
To use Seedance 2, open Dola or the BytePlus Playground, two of the entry points ByteDance links, and describe the shot. Seedance 2.0 also takes reference images, video and audio. ByteDance says 2.0 makes 15-second clips with sound and 2.5 reaches 30 seconds. Seedance makes single clips, so cutting several together, adding captions and exporting needs a timeline editor.
What Seedance 2 takes in and puts out
ByteDance Seed announced Seedance 2.0 on 12 February 2026. Its launch post says you can feed it text, images, audio and video, mixed together. The stated limits are up to 9 images, 3 video clips and 3 audio clips, plus plain-language instructions.
What comes out is 15 seconds of multi-shot video with two-channel audio: background music, ambient effects and character voices aligned to the picture. ByteDance also lists reference-based control of composition, motion, camera movement and effects, editing of chosen clips, and extending a shot.
Seedance 2.5 has its own ByteDance page. It says one generation can run up to 30 seconds and can be extended twice, and it can study a reference video's framing.
| Version | Clip length | Inputs named |
|---|---|---|
| Seedance 2.0 | 15 seconds, multi-shot, stereo audio | Text, image, audio and video; up to 9 images, 3 videos and 3 audio clips |
| Seedance 2.5 | Up to 30 seconds per generation, extendable twice | Reference counts are not stated on the page |
| Seedance 2.0 Fast | Not described on the ByteDance pages we opened | Read the model notes when Woven lists it |
Where to try Seedance 2
ByteDance's Seed pages give three entry points: Try Dola, which opens a chat at dola.com, Try in Playground, which opens the BytePlus playground, and Get API for developers. Those pages list no regions or prices and give no step-by-step instructions.
One rule is stated. Using a real person's portrait as a subject reference needs identity verification or prior legal authorization. Use your own face, or people who have agreed, before you upload anyone's photo.
Write a prompt that gets the clip you want
ByteDance says Seedance can take composition, motion, camera movement, effects and audio from the files you give it. So say what each file is for. A reference that your prompt never mentions leaves the model guessing.
- One subject, one action. A prompt about a barista pouring latte art gives the model one job. A prompt with five events gives you a muddle in the middle.
- Name the camera. Write the move in plain words: slow push-in, handheld, locked-off wide.
- Name the sound. Seedance makes audio with the picture, so say what you want to hear, such as room tone, one line of dialogue or no music.
- Plan for beats. A 30-second take holds several moments in one file. Decide where you will cut before you generate, and ask for the moments in order.
What Seedance does not do
Seedance edits inside a clip. ByteDance lists changes to chosen clips, characters and actions, and extension of a shot. It is not a timeline. It does not join six clips into one video, trim the dead second at the start of each, burn in captions, mix music under dialogue or export for a feed.
That is the work after the clip. Clips from different prompts rarely match in lighting or pace, so cut between them instead of hoping they flow as one take. And generated sound can clash with music you add later.
Use Seedance inside Woven
Woven lists Seedance 2.5 and Seedance 2.0 Fast, not plain 2.0. Both show text-to-video, image-to-video and reference-to-video. The editor has no generate button, so you start in chat: Woven's AI agent reads the live model list and runs the job in the background. Generation uses Woven credits, so check the pricing page if cost matters.
Ask in chat
- Generate a Seedance 2.5 clip of a barista pouring latte art, slow push-in, eight seconds.
- Use this product photo as a reference and make a Seedance 2.0 Fast clip of it turning on a marble counter.
- Make two Seedance takes of the same shot so I can pick the better one.
Or do it by hand
- 1Ask by name. Say Seedance 2.5 or Seedance 2.0 Fast and describe the shot. The agent matches the name to a model on the live list and starts the job.
- 2Keep working. The job runs in the background. When it finishes, the clip appears in chat and is saved in the generated folder of your workspace.
- 3Bring it in. Drag the new clip from the sidebar to the timeline, or open Add, choose Video and select it in the generated folder.
- 4Trim and split. Pull an edge inward to drop the slow start, or park the playhead and press ⌘B. Woven has no ripple delete, so after you remove a piece, slide the next clip back against the one before.
- 5Finish. Choose Add, then Generate captions if the clip has speech. Press M to mute a clip whose sound you do not want, then choose Export and pick 1080p or 1440p.
Common questions
Open Dola or the BytePlus Playground from ByteDance's Seed page, describe one shot and say what each reference file is for. In Woven, ask in chat for Seedance 2.5 or Seedance 2.0 Fast and the clip is saved to your workspace.
ByteDance states 15 seconds for 2.0, with up to 9 images, 3 videos and 3 audio clips as references. For 2.5 it states up to 30 seconds per generation, extendable twice, and gives no reference counts.
The makers' pages do not rank them. Both describe clips with generated sound, 15 seconds for Seedance 2.0 and up to 15 for Kling 3.0. Both list video references, and Seedance 2.0 also lists audio references. The reliable test is the same prompt through both; in Woven both are on one model list.
Yes. ByteDance says Seedance 2.0 outputs two-channel audio covering music, ambient effects and character voices, aligned to the picture.
Inside Seedance you can change chosen clips, characters and actions or extend a shot, per ByteDance. To join clips, add captions and music, and export a finished video, use a timeline editor such as Woven.
ByteDance says real human portraits used as subject references need identity verification or prior legal authorization. Use your own image or one you have clear permission to use.
Sources
- Official launch of Seedance 2.0 - ByteDance Seed, checked October 20, 2026
- Seedance 2.0 - ByteDance Seed, checked October 20, 2026
- Seedance 2.5 - ByteDance Seed, checked October 20, 2026
- AI Video Generator - Kling AI, checked October 20, 2026
- Kling Video 3.0 - Kling AI, checked October 20, 2026
Try Woven free for 3 days
Then $25/mo billed yearly ($299/yr) or $39/month. Native Mac app for short-form video.
Paste into Claude Code to install and connect Woven.