How to use Seedance 2.5 for video-to-video (version 3)?
The first two notes in this series started from words alone: one clip from one prompt, a longer video built in stages. Video-to-video starts somewhere else: from clips that already exist. You upload the clips, and Seedance extends them or stitches them into one continuous journey.
What you need before you start
- An account on Jimeng (free to start).
- Two or more short clips to connect — your own footage, or clips you generated earlier with text-to-video (version 1). In this note, the three clips were made with Seedance itself, for demonstration.
The @Video idea
When you upload a clip, it gets a name: @Video 1, then
@Video 2, @Video 3, and so on. In your prompt you refer
to the clips by those names and describe, in plain words, how the camera should
travel from one into the next. The prompt can be surprisingly short — the
clips themselves carry most of the visual information, so you no longer need the
six-part formula from version 1.
A worked example
Three clips: a dusk flight over a city towards a lit window, the interior of an art gallery, and a painting on the gallery wall. One sentence connects them:
Extend @Video 1; after the window opens, enter the art gallery interior
in @Video 2; finally, the shot moves into the painting in @Video 3.
The result is a single, unbroken camera move: the camera glides through the window at the end of the first clip, travels through the gallery in the second, and finishes inside the painting in the third. Three separate clips become one shot.
Watch the demo video
Open in Google Drive (no account needed).
The video shows the whole process: the three clips are uploaded, the
one-sentence prompt refers to them as @Video 1,
@Video 2 and @Video 3, and the joined video comes
back as one continuous shot.
Watch the example videos
Here are the three clips — @Video 1, @Video 2
and @Video 3 — and the joined video Seedance made from them.
Download the three clips and try the chain yourself.
@Video 1 — the city flight
@Video 2 — the art gallery
@Video 3 — the painting
The joined result
Open in Google Drive (no account needed).
One continuous shot: the camera flies over the city at dusk, slips through the lit window, crosses the gallery, and finishes inside the painting — exactly the journey the one-sentence prompt described.
What makes a good chain
- Each clip should pick up where the previous one visually ends. A window leads indoors; a painting hangs on a gallery wall. The more natural the connection, the more believable the joined video.
- Say the order clearly. Use “after”, “then”, “finally” so the model knows which clip comes first, second and third.
- Keep the prompt small. Describe the journey between the clips, not everything inside them — the clips already show that.
Good to know
- The clips you connect can themselves be AI-generated, so you can grow a long video piece by piece: generate a clip, extend it, then connect the next.
- Official documentation for this feature is on the Volcano Engine site (in Chinese).
Further reading
- Video-to-video reference — the official Volcano Engine documentation (in Chinese).
- Seedance 2.5 prompt guide — ByteDance’s official prompting guide.
- Jimeng — the website where you use Seedance.