Why do AI actors rush their lines?

Updated 2026-10-27·3 min read·Dialogue & audio

The take that talked too fast

In one take from our episode Marry My Broken Son, three dialogue lines were crammed into a four-second beat. The actors raced through them, and it sounded like a recording on fast-forward.

The warning under the beat had already said so: six seconds of speech in a four-second shot.

Why it happens

A video model with native audio has a fixed clip length. If you give it more words than it has time for, it has two choices. It can speak faster, or stop before the line ends. Either way you lose the performance.

It's easy to do by accident. You write a beat, it feels short, so you add one more line, and then another. The text looks fine on the page, but there isn't enough time in the shot.

A rough rule: read the lines out loud at the pace you want, with the pauses. That's the time the shot needs. If the shot is shorter, you need to change the shot or the line.

The fix

1. Read the warning. When a beat has more speech than seconds, it's flagged. Lengthen the beat, or shorten the line. If neither works, split it up.
2. Keep the action and trim the rest. Our take had one thing worth keeping: a champagne throw. We kept that and trimmed the talking off the end.
3. Put each line underneath as its own dialogue line. Every line that came out of the take goes under it, one per row.
4. Build one shot per line. Press Shoot with AI on each line. It builds one shot for that line, framed on whoever speaks it.
5. Set how each line is said. On a single dialogue line, Express writes the emotion tag for you and voices it. One of ours came back tagged gently.
6. Listen before you render. Press Play all to hear the takes first. Typos are spoken too, so check the words.

Why one line per shot works

It gives every line the time it needs. It also gives you a cut on each speaker, which is how a dramatic scene is normally edited anyway. The audience sees who's talking and has time to take in the reaction.

It also makes fixes cheaper. If one line comes out wrong, you only redo that one shot.

Delivery is half the problem

A line that has room can still sound flat. In voice lines, a tag in brackets sets the delivery, like yells or angry. Express does it for you on a single line. Pick the emotion the moment needs, then listen to it.

Quick checks before you render

• Count the lines in each beat. More than one line in a short beat is a warning sign.
• Read them aloud with a timer.
• If a beat is doing an action and a conversation, keep the action and move the talking to its own shots.

If the take is fine but goes on too long, see what to do when an AI character acts too early. For mouth movement, see AI video lip sync.

See the whole episode built in the full guide, and this exact fix in the Short My AI actors rushed their lines.