Who decides where the cuts go in a minute of picture

One page returns up to two minutes inside a single request, in six second increments. The other returns five to fifteen continuous seconds per call. A minute is one job on the first and four on the second, and the difference is who chooses where the cuts fall. As of 2026-09-22.

Inclusion rule. Published length figures only, with the request count worked out by division. No claim is made about how either set of cuts looks. Order. The count first, then what it implies about control.

1Fewer joins is not automatically better

Four requests means three seams to reconcile and three chances for a face or a key light to drift. One request means none of that work exists. For a schedule, the first row plainly wins.

For an advertisement with a fixed beat at the end, it can lose. The cuts inside that one request were placed by the model, so the last six seconds are wherever the model decided they start, and a brief that needs a product shot to land on a particular word has no handle on it.

2The grain matters more than the ceiling

Six second increments mean a slot has to be expressed in sixes. A thirty second window is five beats exactly; a five to six second bumper is one beat and nothing is wasted. An eight second idea has no legal value at all.

On the other row any whole value from five to fifteen is available, so the beat is written first. The cost is that the fifteen second ceiling puts four seams into a minute, and the vendor charges per second of output on top.

3What a plan should read first on each row

On the increment row, read the model identifier condition: anything past the first six seconds requires a particular model ID, and a request against the wrong one comes back short with nothing that mentions versions.

On the continuous row, read the billing basis. Length is charged per second, so iterating at the top of the range costs three times iterating at the bottom, and the cheap order of work is to settle framing short and raise the length once.

4The rest of the column behind these two

Inclusion rule. Every generator this register carries, on the one column this pair disagrees about. Rows that state nothing keep their place and are marked hollow. Order. Alphabetical by generator name.

Why these two. Both rows can fill a minute. One takes the joining work away and takes the placement of the cuts with it; the other leaves both with the editor. That trade is invisible in the length column and decides how the job is scheduled.

• Video duration
  6 second increments, up to two minutesfrom the model tableAmazon Nova user guide, generating videos / recorded 2026-09-22
• Gen-4.5
  Durations are 5-15 seconds on Gen-4.5, for text-to-video and image-to-video alikestated per endpointRunway API, input parameters / recorded 2026-09-22

5Sources

Read from the Amazon Nova user guide and the Runway API input reference, both checked 2026-09-22. The column is described on duration, and each entry has its own cell page at Amazon Nova Reel and Runway, and the rule that keeps surfaces and generators apart is on method. Nearby pairs: CogVideoX and Kling AI, HeyGen and Veo.