How Many Seconds Are in a Second
One. But nobody searches this wanting the number — they're setting a clip length and the field isn't behaving. Duration in AI video is model-specific, not free.
TL;DR
One second is one second. What matters in practice is that clip duration is set per model and is not a free number: some models expose a slider over a range, others only a fixed list of values. Whole seconds only, with no fractional input and no frame rate control anywhere. The floor is 1 second on Grok Imagine Video 1.5 and the ceiling is 30 on Wan 3.0, Wan 3.0 Prime and Seedance 2.5, while most models sit between 4 and 15.
One.
That is the whole answer, and if that is what you came for you can stop reading.
But almost nobody searches this phrase wanting the number. They are setting a clip length in a video tool, the field is not behaving the way they expected, and they have gone looking for the unit. So here is the part that actually matters: in AI video, duration is not a number you choose. It is a constraint the model imposes on you, and it differs on every one.
Duration is per-model, and it is not free
There are two shapes it takes.
A slider with a range. Any whole number between a minimum and a maximum. Seedance 2.0, 2.0 Fast and 2.0 Mini run 4 to 15 seconds. Kling 3.0 runs 3 to 15. Wan 2.7 runs 2 to 15.
A fixed list. No range, just options. Veo 3.1 offers 4, 6 or 8 seconds and nothing between them. The older Kling models offer 5 or 10. Veo 3.1 Lite offers 8, and that is the only value it has.
This catches people out constantly. If you have decided a shot needs seven seconds and you are on Veo 3.1, you cannot have seven seconds. You take six or you take eight, or you change model.
The extremes
Shortest. Grok Imagine Video 1.5 goes down to 1 second, and it is the only one that does. Most floor out at 4 or 5.
Longest. Wan 3.0 and Wan 3.0 Prime reach 30 seconds, as does Seedance 2.5. Most stop at 15.
If clip length is the deciding factor for a shot, pick the model from the length rather than picking the model first and discovering the ceiling afterwards. The model catalog lists what each one takes.
No fractional seconds anywhere
Whole numbers only, on every model. There is no 2.5-second option, no way to type a decimal, no frame-level input.
If you need a clip to land at a precise sub-second duration, generate the nearest whole second and trim in an editor. That is an editing job and always was.
fps is not yours to set
Worth stating plainly, because a lot of writing about this assumes otherwise: there is no frame rate control. No fps field, no dropdown, no way to request 24 versus 30. The model outputs what it outputs.
Which also means the arithmetic people go looking for — seconds multiplied by frame rate equals frame count — is not something you can act on here. You cannot set the frame rate, so you cannot use it to plan.
Related, and it comes up constantly: there is no frame interpolation. No route, no model, no capability. Nothing converts one frame rate to another.
Longer is not more of the same shot
This is the real trap, and it is a creative one rather than a technical one.
Asking for 15 seconds instead of 5 does not give you the same shot held three times as long. It gives the model fifteen seconds to fill, and it fills them — with more camera movement, more action, more happening. Often things you did not ask for.
Short durations are more controllable. One action, a few seconds, is a request a model can satisfy precisely. Fifteen seconds is an invitation to improvise.
Several short clips beat one long one for most work. You cut between them, you keep what worked, and a failure costs you one clip rather than the whole run.
Cost scales with length, so a long generation that misses is an expensive miss. Every generation is quoted before it runs, so the number is in front of you before you commit.
If you genuinely need length
For something continuous and long, generating one enormous clip is the wrong approach anyway. The Long Video Generator chains segments, each continuing from the previous one's final frame, which holds the world stable across a run that no single generation could cover.
For a slow, deliberate movement rather than a long one, first-to-last frame on Seedance 2.0 or 2.0 Fast constrains the distance travelled instead of the speed.
The short version
One second is one second. The useful facts are that duration is model-specific, whole numbers only, somewhere between 1 and 30 depending on which model, and sometimes a fixed list rather than a range.
Pick length first if length matters, keep clips short by default, and cut rather than generate long.
Start in Text to Video, or image to video from a still.
Frequently Asked Questions
How many seconds are in a second?▾
One. Most people searching this are actually setting a clip length in a video tool and want to know how the duration field behaves, which is a more interesting question: duration is set per model, accepts whole numbers only, and ranges from 1 to 30 seconds depending on which model you are on.
Why can't I enter the duration I want?▾
Because some models offer a fixed list rather than a range. Veo 3.1 gives you 4, 6 or 8 seconds and nothing between them, the older Kling models give you 5 or 10, and Veo 3.1 Lite has only 8. If you want seven seconds on Veo 3.1 you cannot have it, so you take six or eight or change model.
What is the shortest and longest clip I can generate?▾
The floor is 1 second on Grok Imagine Video 1.5, the only model that goes that low, since most start at 4 or 5. The ceiling is 30 seconds on Wan 3.0, Wan 3.0 Prime and Seedance 2.5, while most stop at 15. If clip length decides a shot, choose the model from the length rather than the other way round.
Can I enter a fractional duration like 2.5 seconds?▾
No. Every model takes whole numbers only, with no decimal input and no frame-level control. For a precise sub-second length, generate the nearest whole second and trim in an editor, which is an editing job rather than a generation one.
Can I choose the frame rate?▾
No. There is no fps field, dropdown or setting anywhere, so the model outputs what it outputs. That also means the seconds-times-frame-rate arithmetic people go looking for is not something you can plan around here, and there is no frame interpolation to convert between rates afterwards.
Does asking for a longer duration give me the same shot for longer?▾
No, and this is the real trap. Requesting 15 seconds instead of 5 gives the model fifteen seconds to fill, and it fills them with more camera movement and more action, often things you did not ask for. Short durations are far more controllable, and several short clips cut together beat one long generation for most work.
What should I do when I genuinely need a long piece?▾
Chain rather than stretch. The Long Video Generator runs segments where each continues from the previous one's final frame, which holds the world stable across a length no single generation could cover. For slow deliberate movement rather than length, first-to-last frame on Seedance 2.0 or 2.0 Fast constrains distance instead of speed.
