What is Segment vs shot? Definition, key points and related tools
A segment is one video-model call, usually 10–15 s (up to 30 s on Wan 3.0 and Seedance 2.5); a shot is a camera/framing change inside it, 3–6 per segment. Rule: long segments, short shots — fragmenting segments doubles cost, long shots drag.
Two units, two budgets
A segment is the unit the model sees and bills. Each call gets its own text, reference images and voice, and returns one continuous file; nothing carries over to the next call except what you send again. A shot is the unit the viewer sees: one framing, ended by a cut. The two used to be the same thing, one call per shot. Many 2026 models now cut inside a call: Kling VIDEO 3.0 plans up to six shots in one 3–15-second generation (Kling, 2026-07-28), and ByteDance says Seedance 2.0 follows a storyboard's composition, transitions and dialogue placement closely (BytePlus, vendor claim).
The distinction matters because the two boundaries behave differently. Inside one call the model renders one version of the cast, the light and the voices, so a cut between shots keeps them. Between calls it starts again from references and text, so every segment boundary is a place where a face, a timbre or a shadow can shift. Segments set cost and continuity; shots set rhythm.
Segment length is set by the model
| Model | Seconds per call | Cuts inside one call | Source |
|---|---|---|---|
| Seedance 2.0 | 4–15 | Storyboard prompts with transitions (vendor claim) | BytePlus |
| MiniMax H3 | 4–15 (H3-Max 5–15) | Multi-shot prompts (model page) | MiniMax |
| Wan 3.0 | 2–30 | Multi-shot storyboard with timestamps, 4–6 s per shot (Alibaba guide) | Alibaba Cloud |
| Kling VIDEO 3.0 | 3–15 | Up to 6 shots (Multi-Shot) | Kling |
| Veo 3.1 | 4, 6 or 8; +7 per extension | Timestamped shots inside one clip (Google prompting guide) | |
| Sora 2 (API) | 4, 8 or 12 | Separate shot blocks in one prompt (OpenAI guide) | OpenAI |
Fixed steps matter. A 10-second scene becomes a 12-second generation on Sora's 4/8/12, and two calls or an extension on Veo's 4/6/8; on range-based models (4–15, 2–30) the segment can match the scene. Pipeline tools such as SceneMixer make segment length a project setting and plan the cuts inside each segment for you; on a model platform you choose length and cuts per call, which means more decisions and more control.
Where to put the boundaries
- Break segments where the story cuts anyway: a change of location or time, a new scene, an exit. A hard cut is expected there, so a small shift in light or framing reads as editing.
- Keep a dialogue exchange inside one segment. Splitting a back-and-forth across two calls re-renders both voices and both faces mid-conversation.
- Plan 3–6 shots in a 10–15-second segment, each sized by its content: dialogue shots by the words they carry (dialogue timing), silent shots by the action they show.
- Where a model has no multi-shot format, plan one call per shot and carry continuity in the text: each call's opening state restates the previous call's end state (storyboard script).
Common mistakes and their causes
- Fragmenting: one call per shot. Per-second billing prices the seconds the same either way; what multiplies with the number of calls is everything else. Per-call surcharges repeat (MiniMax H3 charges $0.04 for each reference image beyond five in a task, MiniMax pricing), fixed duration steps round up, every boundary is a new chance of drift (and of a retake), and there are more files to review. So fragmenting raises the bill even though the price per second is unchanged.
- Over-long segments. The opposite trade: fewer boundaries, costlier retakes. One wrong shot in a 30-second Wan segment means regenerating all 30 seconds, and OpenAI's Sora guide notes that its model follows instructions more reliably in shorter clips, suggesting two stitched 4-second clips over one 8-second clip (OpenAI). On the drama engines above, 10–15 seconds is the working middle.
- Long shots. A six-second static shot carrying one line drags on a phone. The fix is more cuts inside the segment, not more segments.
- More shots than the format holds. Kling's multi-shot mode stops at six, and Alibaba suggests 4–6 seconds per shot on Wan 3.0, so a 12-second Wan segment holds two or three shots, not six. A storyboard with more shots than the format holds leaves the model to decide what to merge.
Key figures and sources
| Figure | Value | Source | Date |
|---|---|---|---|
| Seedance 2.0 seconds per call | 4–15 s | ai.byteplus.com | 2026-09-12 |
| Wan 3.0 seconds per call | 2–30 s (input plus output video ≤ 30 s) | www.alibabacloud.com | 2026-09-12 |
| Kling VIDEO 3.0 multi-shot | Up to 6 shots in one 3–15 s generation | kling.ai | 2026-09-12 |
| Veo 3.1 clip lengths | 4, 6 or 8 s; each extension adds 7 s | ai.google.dev | 2026-09-12 |
| Sora 2 API clip lengths | 4, 8 or 12 s | developers.openai.com | 2026-09-12 |
FAQ
How long should a segment be?
As long as the scene allows within the model's range, usually 10–15 seconds: long enough that a conversation stays inside one call, short enough that one bad shot does not force a long retake. Wan 3.0 accepts up to 30 seconds; Veo 3.1 and Sora 2 use fixed steps (4/6/8 and 4/8/12 seconds).
How many shots fit in one segment?
Three to six in 10–15 seconds is the usual range, and Kling VIDEO 3.0 caps its multi-shot mode at six. Vendor guidance differs: Alibaba suggests 4–6 seconds per shot on Wan 3.0, which means fewer, longer shots per segment.
Does a longer segment cost more?
Per second, no: these engines price by output seconds, or by tokens or credits that scale with them. Per retake, yes: regenerating a segment re-bills all of its seconds, so one bad shot in a 30-second segment costs twice as much to fix as the same shot in a 15-second one.
Sources
- https://kling.ai/blog/kling-video-3-multi-shot-guide
- https://kling.ai/quickstart/klingai-video-3-model-user-guide
- https://www.byteplus.com/en/blog/dreamina-seedance2-0
- https://ai.byteplus.com/en/activity/seedance2-0
- https://platform.minimax.io/docs/api-reference/video-generation-v2-create
- https://platform.minimax.io/docs/guides/pricing-paygo
- https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-api-reference
- https://ai.google.dev/gemini-api/docs/veo
- https://developers.openai.com/api/docs/api-reference/videos/create
- https://developers.openai.com/cookbook/examples/sora/sora2_prompting_guide
- https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1
- https://www.alibabacloud.com/help/en/model-studio/wan3-video-generation-guide