Write shots, not one giant prompt
Give every segment one clear action and one camera intention. Use the cut to change location, scale or point of view.
MiniMax H3 long video calculator
H3 makes a short shot feel immediate. The interesting opportunity begins when those shots become a scene, a channel or a small film. This planner prices the joins, retries and references that a simple seconds-times-rate calculation misses.
Build the sequence
The expensive part of a long H3 video is not the final runtime. It is the overlap, discarded tail and extra attempts needed to make the joins disappear.
Reference inputs are provider- and workflow-specific, so they stay editable instead of being guessed.
Assembly map
Overlap is editorial breathing room. Keep it only while matching motion, color and sound across the cut.
| Clip | Generate | Keep | Editorial job |
|---|---|---|---|
| 01 | 0–15s | 0–15s | Lock the opening look and sound bed |
| 02 | 14–29s | 15–29s | Blend the first 1s into the previous cut |
| 03 | 28–43s | 29–43s | Blend the first 1s into the previous cut |
| 04 | 42–57s | 43–57s | Blend the first 1s into the previous cut |
| 05 | 56–71s | 57–60s | Blend 1s, then trim at 60s |
Why H3 matters
The old unit was one expensive finished video. With H3 Max, the useful unit is an accepted shot: picture, motion and sound tested together, then kept or replaced quickly. That makes ambitious formats feel reachable to people who do not have a studio behind them.
Long-form work still needs taste. Continuity has to be designed, dialogue has to survive the cut, and the best take must be chosen. The model removes waiting; it does not remove directing. That is exactly why the opportunity is interesting.
A workable sequence
Give every segment one clear action and one camera intention. Use the cut to change location, scale or point of view.
Carry forward the few details viewers notice: face, wardrobe, light direction, lens language and the final pose.
Let ambience bridge the edit. Avoid restarting music, room tone or dialogue cadence in every generated clip.
Build from an accepted segment. Continuing from a weak shot only makes the continuity problem more expensive.
The planning model
The planner calculates the number of fixed-length clips needed to cover the target runtime after seam overlap. It then adds the unused tail, an expected retry reserve and the optional cost of each continuation reference pass.
total = generated seconds × output rate × retry multiplier + reference passesBefore the first shot
H3 workflows normally generate short clips. A longer piece is assembled from planned segments, using accepted frames or reference media to carry the look and action into the next shot.
A credible budget includes overlap for clean joins, unused tail on the final segment, retries, and any reference-input charges required by the chosen endpoint.
Start with a small overlap and increase it only when the action or audio needs more room to blend. This planner defaults to one second and keeps the assumption editable.
Yes, through an editable cost-per-continuation field. Reference billing differs by endpoint and input pack, so the planner does not hide a guessed fee inside the output rate.