30-second single pass versus stitched 8-second shots
Seedance 2.5 can do ~30s in one pass. Veo still thinks in ~8s with extension. Stitching is a continuity tax.
One pass when identity must hold. Stitch when you need different rooms.
That is the production rule. The spec for what Seedance 2.5 actually emits — native ~30 seconds, one denoise, audio in the same pass — lives at Seedance 2.5: 30-second single pass. This page is the decision that spec does not make: when a 30-second take is the cheaper continuity, and when four 8-second shots are the cheaper control.
Stitching is a continuity tax. You pay it in identity drift, in room-tone seams, in the extra stills you have to lock so the cut does not recast the lead. You do not pay it when the cut is the point.
What each clock actually is
Seedance 2.5 generates 4–30 seconds in one pass, at up to 720p, with native audio. The model reasons about the whole shot at once. Motion, light, and subject are supposed to still agree at second 28 with second 2. That is the claim you are buying. You are not buying 4K. You are not buying a cut.
Veo 3.1 still thinks in 4 / 6 / 8 seconds per generate. Longer is extension: a fresh 7-second beat that continues the last frame, not a 30-second brain. Always-on dialogue and a 4K tier are the reasons to live on that clock. Duration is not.
Stitched 8-second shots are any model, including Seedance, cut together in an editor. Each generate is its own identity event and its own audio event. A hard cut hides a lot. A pretended single take hides nothing.
Scene chaining is the honest version of stitching: last frame of clip N becomes the I2V input of clip N+1. That is a tax you choose because you need a new camera setup, not because you wanted a longer take of the same one.
The decision
Ask what has to be true at the end of the clip.
Identity must hold across one continuous action, in one room, with one soundtrack. Single pass. A 20-second pour. A presenter who cannot recast at second 12. A walk that has to be one walk. If you stitch that, you are betting the cut is invisible and that two independently sampled rooms will sound like one room. They will not.
You need different rooms, different lenses, or a pack shot that must match a still. Stitch. A three-setup ad (kitchen, bathroom, pack) is not a 30-second problem. It is a stills problem. Lock the stills, animate each setup, cut. A single 30-second pass will invent the bathroom. It will not honour the pack.
The mouth is the shot and the line is 8 seconds. Veo, on Veo's clock. Do not put a talking-head close-up through Seedance 2.5 just because 30 is a bigger number. Do not put a 25-second in-world speech through Veo and extend four times because you like the voice. Match the clock to the mouth's duration, then to the room.
| Brief | Clock | Why |
|---|---|---|
| One action, one space, 15–30s, identity cannot recast | Seedance 2.5 single pass | Continuity is the deliverable |
| Talking head, always-on dialogue, ≤8s | Veo 3.1 | Audio job, not a duration job |
| Talking head that must keep going | Veo + extend, or a talking-shot specialist | Extension continues a take; it does not become a 30s brain |
| Three locations, one product | Stitched I2V from locked stills | Each room is a still. The cut is a feature |
| 30s ad that is really a hook + body + CTA | Stitch at the structural cuts | Those are edits. Do not ask a single pass to "cut" |
| Landscape / weather / abstract that can wander | Single pass, cheaper model | Identity is not on the line |
The tax, itemised
People stitch because 8-second models are familiar and because an 8-second keeper feels cheaper than a 30-second roll. Count the whole file.
- Identity retries. Four 8-second T2V shots of "the same woman" are four castings. A single pass is one. If you already have a locked still, I2V stitch is cheaper than T2V stitch. It is still more identity events than one pass.
- Audio seams. Native audio on clip A and native audio on clip B are two rooms. A cut can hide a picture jump. It almost never hides a tone jump. One pass has one room.
- Handoff frames. Chaining needs a settled last frame. A generation that ends in motion blur poisons the next shot. A single pass has no handoff.
- Resolution and grade. Seedance 2.5 tops out at 720p in the catalog. If the deliverable is 4K, Veo's clock may be the right tax. Do not single-pass a hero you will later be told to upscale out of a 720p plate unless that is an accepted trade.
The 30-second number is also a trap the other way. Hitting the number is not the same as needing it. A 12-second idea padded to 30 is a retention problem, not a model problem. Size the pass to the action.
What not to do
Do not stitch four arena-winning 8-second clips and call it a Seedance competitor. You made a montage. Say so.
Do not single-pass a film that needs three rooms to "avoid editing." You will spend the continuity you bought on a bathroom the model invented, then cut anyway.
Do not extend Veo four times to fake a 30-second take of a product that has to match a pack shot. Extension continues whatever the last frame is, including its errors. A pack shot is a still. Animate the still.
Do not pick Seedance 2.5 for a 6-second hook because the model can do 30. Short is still a first-frame job. Use a short clock.
FAQ
Is a Seedance 2.5 30-second generate always better continuity than chained Veo?
For one action in one room, yes — one denoise, one identity, one soundtrack. For a sequence that has to change lens or location, chained (or separately still-locked) shots are better control. Continuity of a person is not the same as continuity of a floor plan.
Can I single-pass the body and stitch a hook in front?
Yes. That is the honest hybrid: a 3-second opening you are testing, and a body that must not recast. Generate the body as one pass (or one locked I2V). Do not regenerate the body to "match the hook's energy." Matching is how the test dies.
Why not always generate 30 and trim?
Because a 30-second pass that is really an 8-second idea will wander. The model fills the time. Trimming a wander is slower than generating the length you meant. Also, you paid for the whole pass.
Does extension count as a single pass?
No. Extension is a new generate conditioned on the last state. It is the right tool when the last frame is good and you want more of the same shot. It is not Seedance's 30-second brain. If you need 30 seconds of held identity with evolving audio, that is the single-pass job. If you need 8 more seconds of a take you already like, that is extend.