DEV Community

Cover image for How I Changed One Line of Dialogue with MiniMax H3
Joey
Joey

Posted on

How I Changed One Line of Dialogue with MiniMax H3

Carrie turns away from the sofa, takes a few steps, and looks into the camera. Then she says, “Let’s begin.”

I liked the shot. I just wanted her to say “Ready when you are.”

My first instinct was to change that line in the original prompt and run it again. The new line worked, but Carrie moved differently. I’d fixed the dialogue and given myself a new performance to review.

Give the model the shot you want to keep

The tricky part was that Carrie’s face was visible. Swapping the audio would give me the right words, but her mouth would still be saying the old ones.

So I generated “Ready when you are” as a separate audio clip. Then I connected that audio and the original video to a new MiniMax H3 video node. You can see how I wired the two inputs in my Astorie Canvas walkthrough.

The inputs had different jobs. The video showed H3 how Carrie moved and how the camera approached her. The audio gave it the new line and its timing. I kept the instruction focused on the edit:

Use the source video as the visual reference. Replace the spoken line with the supplied audio and adjust the mouth movement to match. Keep the body motion, framing, and camera movement as close to the source as possible.

Watch the lead-in, too

In my six-second clip, Carrie’s mouth moved with the new line. Watching the two versions, I didn’t see a meaningful change in her body movement, the room, or the camera push-in.

H3 still generated a new clip, so I checked the few seconds before she spoke as well. That’s where I could see whether the turn and the camera move I already liked had carried through the edit.

Top comments (0)