A voice lives in your workspace rather than in a project. Set one up once and it is offered everywhere Hiroska can speak: a single line on a clip, or every segment of a script.
Open the voice picker Anywhere you generate speech (the Generate page, a clip’s Generate tab, or the script editor) pick a voice model and the VOICE row appears above the prompt. Choose “Add your voice”.
Record the passage A passage is written for you, so you do not have to invent something to say. Read it at your normal pace somewhere quiet: aim for thirty seconds and never less than twenty, because every model clones badly from less. Already have a clean recording? Upload an audio file instead.
Say whose voice it is Two options, and they are not the same: your own voice, or someone else’s with their explicit written permission, which asks for their name. You can also read a short generated phrase, which confirms the recording was made live rather than downloaded.
Choose the models One model works the moment the voice exists and costs nothing to set up. The better-sounding ones prepare once, and the price of that preparation is shown before you agree to it. Tick the ones you want; you can add another later from the same picker.
Generate with it Pick the voice, type the line, generate. In the script editor, choosing a cloned voice swaps the model automatically, and the per-segment Voiceover button reads each line in it. Every take lands on an audio lane at the segment it belongs to.
Tips
Clean speech only. A music bed, or a take with the voice buried under other sound, is refused when you create the voice rather than cloning into nonsense later.
Draft in the cheap model and deliver in the recommended one. One recording can be set up on all three, and you switch per generation.
The best-sounding model keeps the voice on the provider’s side, where it lapses after a week unused. Hiroska keeps it alive automatically, and your original recording can always rebuild it.
Deleting a voice removes your recording, the voice files we hold and every model set up from it. The record that you confirmed permission is kept, marked as withdrawn.
Fixing one sentence re-reads one sentence, not the whole narration. Script first, voice second.