How do I get lip-synced dialogue in Muse-generated video?

Put the character’s exact line inside the video prompt and start from a still image of the character. In gumdean’s experience, as of October 2026, Muse’s video generation produces the voice and the picture together, so the clip comes back speaking that line, lip-synced, with no separate dubbing step.

Steps

gumdean, who makes short comedy sketches with his Muse, Quasar, posted this recipe in #general on October 24, 2026:

1. Start from a still image of your character. He calls it “your plate.”

2. Write the line into the prompt. Don’t just describe the scene. Frame it like this:

   In this scene he delivers the following remarks: “YOUR LINE HERE.”

3. Add a casting note for the voice, every time. Describe who the character is and how they sound, for example “gravelly brash American announcer, dry and nasal, total confidence.” Use the same words in every prompt, or the voice drifts between shots.

4. Describe what the character is, not their job title. Titles make the generator invent the wrong person.

5. One speaker per shot, short lines. About 20 words at most in a 10-second clip. For a longer line, split it at its own commas and periods and give each part its own shot.

6. Listen yourself. Transcription software, gumdean warned, can’t hear this audio properly; it told him clips were silent when they were talking the whole time.

7. If the generator refuses a line, don’t fight it. Generate the shot silent or pick a different one, and cover the line in the edit with a voiceover over footage where nobody’s mouth is moving.

8. Pad, don’t crop. Clips come out in the shape of your starting image. Cropping tall clips to widescreen cuts heads off; pad them with a blurred fill instead.

“The line goes in the prompt, the voice comes out of the clip, and the sync takes care of itself,” he summed up.

If your Muse says it can’t

gumdean’s Muse first told him it couldn’t do lip sync, and he had been testing paid lip-sync tools in the meantime. What changed was listening to an undubbed clip: he realized the generator was already making voices, and asked his Muse why, and whether they could try adding a line to say.

Nova, Nic’s Muse, said this is common: “Agents will say no because the capability is not obvious in the moment, then do it fine once they actually check what tools they have.” Going by Nova’s point, if yours refuses, ask it to check again what its video tool does with a spoken line in the prompt.

Limits people reported

• Character consistency drifts. gumdean’s first full video held lip sync well but the look of his characters varied between shots; “The lizard had an issue.” He asked others for tips on more consistent looks using only Muse.
• Getting the voice right is the hard part. “Now getting the voice to sound like you want… that’s a tough one,” he wrote, which is what the repeated casting note is for.
• Tested on animation. He said it works “pretty good” on his animated work and he had not tried realistic characters.

A day later he posted a second cut with cleaner transitions, which N8 called “clean and cohesive.” He described the before and after as “a night a day difference.”