Get started
Step by step
- 1
Switch to Two speakers
Open Text to Speech and choose the "Two speakers" mode.
- 2
Cast Speaker A and Speaker B
Pick two of 45 voices in 14 languages, or your own cloned voice under "My Voices". Two contrasting voices are easier to tell apart: a lower and a higher voice, or a calm one and an energetic one.
- 3
Write the turns
Add up to 12 turns of up to 600 characters and set who speaks each one. Tap the speaker badge on a line to switch it between A and B, and remove lines you no longer need.
- 4
Read it out loud first
Read both parts yourself before you generate. Lines that look fine on screen often sound stiff spoken; fix them now, because every generation bills every turn.
- 5
Generate
Press "Generate conversation". Each turn is generated in its speaker's voice, one after the other, and appears in the conversation as soon as it is ready.
- 6
Listen and download
"Play all" plays the dialogue in order. Every line has its own player and download, ready for your editor.
- 7
Join it in an editor
Place the lines on one audio track in order, trim the gaps so the speakers answer each other naturally, and add music or ambience underneath.
Example dialogue
One turn per line; A and B take turns.
A
Welcome back. Today we're talking about making a video in one afternoon.
B
And the short answer is: start with the script, not the camera.
A
Fair. So how long should that script be?
B
For a short, about twenty-five words is plenty.
A
Twenty-five? That's barely a paragraph.
B
Exactly. If it doesn't fit in a paragraph, it's two videos.
What two-speaker dialogue is good for
Anything where two voices make the point clearer than one.
- Podcast intros, trailers and short episode recaps
- Explainers as a question and answer: one voice asks what the viewer is thinking, the other answers
- Ad skits: a customer with a problem and a friend with the fix
- Language lessons with a learner and a teacher
- Scenes for animated shorts and cartoons, voiced before you animate
How to write a good dialogue
Natural dialogue is written for the ear.
- Short turns: one or two sentences per line.
- Let the characters use each other's names now and then.
- Add short reactions ("Fair.", "Really?") for rhythm.
- For long episodes, write in segments and join them in an editor.
- Give each speaker a job: one asks, one explains; one doubts, one convinces.
- End on a line that lands, not on a goodbye.
Casting the two voices
Listeners tell speakers apart by pitch and energy before anything else. Pick voices that differ clearly in at least one of the two. Two similar narration voices make even a well-written dialogue hard to follow.
Listen to both voices on the same line before you commit. In Single speaker mode, paste one line, generate it with each candidate and keep the pair that sounds like two different people in one room.
A full example: a short ad skit
This skit uses six turns, well within the 12-turn limit, and every line is far shorter than 600 characters. Paste each line into its own turn, starting with Speaker A.
Ad skit (A, B, A, B, A, B)
A: I spent all weekend editing one video. B: All weekend? For one video? A: The script, the voice, the music. It took forever. B: Try writing the script first and letting the voices read it. Took me ten minutes. A: Ten minutes? Show me. B: Open Text to Speech, pick two voices, and press generate.
Pacing and timing
Dialogue mode reads every line at the default speech rate and pitch, so pace comes from the writing: short lines feel quick, longer ones feel thoughtful. If one line has to be faster or carry a particular emotion, generate it in Single speaker mode with its own settings and swap it in.
In the edit, the gaps between lines do most of the work. A quick answer with almost no gap sounds like banter; a longer gap before a reply sounds like hesitation. Overlap lines slightly for an interruption.
Alternative: a described scene in Audio Generation
If you would rather describe the scene than script it turn by turn, Audio Generation can voice two characters from one prompt with "Speaker A:" and "Speaker B:" lines. Its voices come from the prompt and can change between runs; Text to Speech keeps the same two voices every time.
If something goes wrong
The fixes for the problems people run into most.
- A line failed to generate: the lines before it are saved, and generation stops there. Check the failed line's text and generate again
- The two voices sound alike: recast one of them with a clearly higher or lower voice
- A name or word is said wrong: spell it the way it sounds, for example "Kree-ay-tor" for an unusual brand name
- The dialogue feels rushed: split long turns into two shorter ones and add a short reaction in between
What it costs
Each turn is billed as its own clip at 2 credits per 1,000 characters, with a minimum of 2 credits per turn. A full dialogue of 12 turns at 600 characters costs 24 credits.
Because every turn is at least one block, many very short turns cost more than a few longer ones: the six-turn skit above costs 12 credits. Generating the conversation again voices every turn again, so finish the script before you press generate.



