RenderDrama

What a second of AI video costs, channel by channel

Counting the joins an episode will need

An episode is assembled from as many generations as its runtime divided by the ceiling on one call. That number is a schedule fact rather than a cost: the footage costs the same either way and the review does not. As of 2026-09-12.

How the piece count moves with the ceilingFifteen minutes of episode divides into about thirty generations at a thirty-second ceiling and a hundred and eighty at five seconds. Every boundary between two generations is where a face drifts or a light jumps, and the footage costs the same at every ceiling.One fifteen-minute episode, at three ceilingsAbout thirtypieces30 secondsA segment canhold a smallscene with aturn in it.About sixtypieces15 secondsA segment is abeat: oneexchange, oneaction, onereaction.About 180pieces5 secondsA segment is asinglemovement, andreview workgrows with theboundaries.The bill follows the seconds; the calendar follows the pieces
Fig. 1 Retakes multiply the generation count and not the join count: a one-in-three rejection rate needs ninety generations for sixty pieces.
How the count of generations and joins moves with the ceiling, for one fifteen-minute episode. Recorded 2026-09-12.
Ceiling on one generationGenerations for fifteen minutesJoins to survive
Thirty secondsAbout thirtyTwenty-nine
Fifteen secondsAbout sixtyFifty-nine
Ten secondsAbout ninetyEighty-nine
Five secondsAbout a hundred and eightyA hundred and seventy-nine

Inclusion rule. Arithmetic on a fifteen-minute runtime at four ceilings. The figures are division, not readings from any page, and they ignore retakes. Order. From the longest ceiling to the shortest.

1Joins are where the failures a viewer notices happen

A face drifting, a light jumping, a costume changing shade: each of those becomes visible at a boundary between two generations. Doubling the number of boundaries roughly doubles the chances of one, and the review work grows with it.

None of that is priced. A per-second rate is indifferent to whether fifteen minutes arrives in thirty pieces or a hundred and eighty, and a production is not.

2A longer ceiling changes what a segment can hold

At fifteen seconds a segment is a beat: one exchange, one action, one reaction. At thirty it can carry a small scene with a turn in it, which removes a boundary and removes the continuity risk that boundary carried.

That makes the ceiling a writing constraint as much as a technical one. A shot list built for a fifteen-second limit and generated against a thirty-second one wastes the headroom, and the reverse does not fit.

3Retakes multiply the count, not the ceiling

Every figure above ignores discarded attempts. At a rejection rate of one in three, a sixty-piece episode needs ninety generations and still has fifty-nine joins. The join count follows the ceiling and the generation count follows the ceiling and the rejection rate together.

So the two numbers to write down before comparing channels are the ceiling and the expected rejection rate. Between them they determine both the bill and the calendar, and only the rate card mentions either.

Written to explain, not to record. No sentence here is attributed to a seller, and none of it should be quoted as a fact about one. The sourced material is on the ledger. Related: Where terms live, Per-second billing.

4Inside a generated second

Price aside, what goes into the seconds behind “Counting joins” is described in SceneMixer's guides:

Paste a scene or a short script and SceneMixer gives you a storyboard — numbered shots with framing and seconds — and then a moving version of each segment, which you can join into a rough cut of the scene.

SceneMixer, Using SceneMixer for Previs and Animatics Before a Live-Action Shoot (read 2026-10-13)

That is SceneMixer's core design: characters live in a persistent asset library with locked reference art, and every segment that features a character is generated against those references — same face in episode 1 and episode 50.

SceneMixer, The AI Micro Drama Production Pipeline — “Can AI keep the same actors across 50 episodes?” (read 2026-10-13)