Two more H3 Max doors: one makes a photo talk, one moves the camera

Lip Sync has no box for words at all. There is nowhere to type, because what gets said comes out of the recording you hand it, and the clip comes back exactly as long as that recording. The sound has to run at least 5 seconds, and anything past 14.8 seconds is cut, so a 20-second line comes back as the first 14.8 seconds of it. The photo has to be between 0.4 and 2.5 times as wide as it is tall, which a normal portrait or a normal landscape both pass.

Camera Controls is the other way round. You type what you want, you pick a length between 5 and 15 seconds, and the picture you hand it is the scene the camera moves through. Nothing in the scene animates on purpose.

Both charge by the second at the same rate, and Lip Sync goes one step higher: it is the first video model here with a 2K step. A new Lip Sync box starts at 768P and a new Camera Controls box starts at 480P, which are the makers’ own defaults.

Every other video model here is handed a whole number, because the length is a control you set and you cannot set half a second. Lip Sync is the first one whose length comes off a file somebody recorded, and a recording is never a whole number of seconds.

So we ran one real generation against the bill: a 7.488-second recording at 480P. The clip came back 7.5 seconds long and MiniMax charged for 8. Our own arithmetic had said 7.488 seconds, which was 6.8% under what we were actually charged, and that is the wrong direction to be wrong in. The rule now rounds the second up, so the number on the button is the number on the bill.

For you that means a 7.5-second recording is paid for as 8 seconds. Nothing is hidden: the Generate button already shows the rounded figure before you press it.

A Camera Controls box draws a row of named moves: Push in, Pull back, Pan left, Pan right, Tilt up, Tilt down, Crane up and around, Orbit the subject, Drone, Handheld, and thirteen more. You pick one and it writes the move into the box.

Three of them also carry real geometry, a list of camera positions over time: push in, pull back, and orbit the subject. The other twenty travel to the model as words, which is how MiniMax drives this model themselves. Words cannot point the wrong way, and that matters here: nobody publishes which direction a positive number means, so a move described with numbers can come back mirrored.