Why creators use it
- Your voice on every video, even when you're not in a place to record: clone it once from a clean sample.
- Faster iteration: try five versions of a hook and keep the one with the best timing.
- Characters and skits with distinct voices, from samples you record in character.
- Series in other languages, in the same voice, for audiences abroad.
Writing for short-form
- The hook is the first line, and the first second. State the payoff or the question immediately.
- Budget the words. At about 150 words a minute, 30 seconds holds roughly 75 words and 60 seconds about 150. Leave room for on-screen moments. See script length and timing.
- One beat per line: each line matches a shot or a caption, so cutting is easy.
- Short sentences, conversational words, contractions. Written for the ear, not the screen.
Narrator: This one setting doubled my battery life.
Narrator: Open settings, then battery.
Narrator: Scroll all the way down. See that switch? Turn it off.
Narrator: That's it. Follow for part two, where it gets weird.How to make one
- A project for your channel, with your voice in the cast and a script per video.
- Record your sample with the energy of your videos. The clone copies it; a sleepy sample makes a sleepy short.
- Choose a style: Balanced for most content, Expressive for high-energy channels.
- Generate the script, and re-take the hook until the timing feels right.
- Download, drop the numbered files on your timeline, and cut the picture to them. See syncing to video.
- Captions: the script text comes in the download, ready to paste into your editor's caption tool.
Sound that survives phone speakers
- Keep music well under the voice, and duck it while you're talking. See mixing with music.
- Check the final video on a phone speaker before you post.
Labels and rules
TikTok, YouTube and Instagram all have policies on AI-generated or synthetic content, including labels to switch on when you upload, and they apply most strongly to realistic content that could mislead. A cloned voice of yourself, clearly presented as you, is the easy case; a cloned voice of anyone else needs their consent and a disclosure. Rules change often, so check each platform's current policy. See disclosing AI voices.