Clone and design voices

Before you start

Authenticate and confirm sufficient quota. Clone only audio you own or are authorized to use. Keep consent and source records. Do not use another person's identity to deceive listeners.

Clone an authorized sample

kitta audio voices clone --schema
kitta audio voices clone --name Narrator --audio-file sample.wav --visibility private

Replace sample.wav with your authorized local recording. This command uploads the file as multipart audioFiles. Inspect the returned JSON for the resulting voice ID, then verify it:

kitta audio voices get --voice VOICE_ID

Design, preview and save

kitta audio voices design --prompt "A calm warm narrator" --preview-text "Hello from Kitta"

Keep the design ID and candidate IDs from the result. Candidates are temporary; design is not the same as saving a reusable voice.

kitta audio voices preview --design DESIGN_ID --candidate CANDIDATE_ID --output preview.wav
kitta audio voices save --design DESIGN_ID --candidate CANDIDATE_ID --name Narrator

Replace both IDs with values from the same design result. Listen to preview.wav before saving. Use the saved voice ID with TTS.

Partial or unknown results

Known limitation observed on 2026-09-22: some voices returned by the list endpoint may return ERR_MODEL_NOT_FOUND from voices get even when synthesis works. A detail 404 alone does not prove the voice is unusable. Preserve the voice ID and request ID for support; do not create a paid clone to work around a discovery inconsistency. Recheck current service behavior before depending on detail lookup in automation.

A partial design result is still returned on stdout, with exit code 5. Preserve it and inspect the available candidates instead of assuming nothing was created. Paid cloning and design requests are not automatically retried. If the outcome is unknown, inspect your account before submitting again. Use account controls to manage unwanted voices; do not invent an unsupported CLI delete command.

Next: recovery and billing.