Runs on a GPU from Modal's free monthly credit (capped at $15, then falls back to your free server: slower, still unlimited). Write numbers as words, and use commas and full stops for pauses.
The model copies the delivery of your recording too, so record in the style you'll use it for. You can make one voice per style.
Anything from 10 seconds to 10 minutes: a reel, a voice note, a class recording. Only your voice, no background music. It's transcribed automatically and the best 8–14 s stretches are picked for you to choose from.
The model uses one 8–14 s clip as the voice. Pick the cleanest one with the delivery you want: no singing, laughing or noise. Fix the text if a word is wrong; it must match what's said exactly.