How to write spoken lines for AI video

A line that reads well can still sound wrong out loud. AI voices take their direction from the text itself: its length, its punctuation and the tone you tag it with. These are the rules Yuumeh uses for every spoken line.

Voices first, video second

Don't find out how a line sounds after you've paid for the video. On Yuumeh, lines are voiced first. You pick each character's voice, hear the lines, fix what's off, and only then is the video made to fit the voices.

1. 1Make the characterStart with a picture, then make a character card from it, so the same face comes back in every shot.
2. 2Pick a voiceListen to voices from the library and choose one for the character.
3. 3Write the linesOne row per line, written the way it's said.
4. 4Hear themGenerate the voices and listen. Change the words, the tone or the punctuation until each line is right.
5. 5Make the videoThe video is made to the voices, so mouths and timing follow the lines you approved.

How to write a line

• Keep it short. Under 14 words. Split anything longer into two lines.
• One row per line, as Name: [tone] "words".
• Give it a tone: calmly, deadpan, whispering, out of breath.
• Punctuate for delivery. The voice reads punctuation: "Fine." is not "Fine…" and neither is "Fine?"
• Put the body between the lines. On the rows in between, say what the character does: she looks away, he puts the cup down. That's where the pauses come from.

Mara: [calmly] "You're early."
She doesn't look up from the ledger.
Jon: [out of breath] "The bridge is closed."
Mara turns a page, slowly.
Mara: [deadpan] "Then we have a problem."

Mistakes to avoid

• A speech on one row. It comes out rushed and flat.
• Lines stacked back to back. With nothing between them, there's no pause and no acting.
• A tone doing the writing's job. No tag can save a line nobody would say out loud. Read it aloud first.

What it costs

Voices are priced by the characters voiced, tone tags included: 16 credits per 1,000 characters. A short exchange costs a few credits, which is why it's worth getting the lines right before the video. See pricing.

Questions

Why does AI dialogue sound robotic?

Usually because the line is too long, has no punctuation to shape it, or runs straight into the next line with no pause. Shorter lines, a tone for each and an action between lines fix most of it.

Can I use my own voice?

Yes, your own or one whose speaker has agreed. Cloning anyone else's voice is against Yuumeh's rules.

Can I make a conversation without video?

Yes. A Dialogue card voices a whole exchange with tones and pauses, for a voice-over, a radio play or a scene to plan around.

Reviewed 11 November 2026