Safety
Cinara can make a convincing copy of a voice, so it asks the speaker first, checks the answer, and keeps a record. Here is exactly how, and where the checks still have limits.
Voice cloning
Every cloned voice starts with the speaker saying, on the recording, that they agree. A tick box alone is never enough.
You enter the speaker's full name and confirm you are that speaker, or have their permission.
Cinara writes a statement for this recording only: “My name is … I agree to let Cinara create a digital copy of my voice. My one-time code is …”, with four random code words. It expires in 30 minutes.
The recording, 20 seconds to 5 minutes, starts with the statement read aloud, followed by natural speech.
All four code words must be heard, in order; at least 80% of the statement must be heard; exactly one person must be speaking; and there must be at least 20 seconds of speech.
The voice is made only after the check passes. If it fails, no voice is made and the credits are refunded. Each statement works for one recording.
Each cloned voice keeps its statement, the code words, how much of the statement was heard, the number of speakers, the length of speech, the confirmation, and who made it and when. The Voice Library shows the date consent was verified.
A singing voice learns from the clone's consent recording and up to 6 more. Each extra recording starts with its own consent statement, which must name the same speaker, and passes the same checks.
Names of well-known people are refused as speakers and voice names in Voice Cloning, in Voice Design descriptions, and in prompts for Images, Video and Talking Avatars.
Music refuses well-known artists in styles and lyric ideas, so songs are written in a style, not as a copy of someone. To hear a song in your singing voice, use songs you made or have the rights to.
Talking Avatars asks you to confirm you have the right to use the photo and voice, and that you won't use the video to impersonate, mislead or harm anyone. Dubbing gives each speaker a matched voice; it never clones the people in your video.
Every account has a verified phone number, one account per phone. Every generation is linked to the account, workspace and API key that made it, and we use that trail to investigate abuse. Images held back by the safety filter are dropped and not charged.
Automated checks and reports flag possible misuse, and the Cinara team looks into reports. We may remove content, turn off links, disable voices or keys, or close accounts. If we act on your account, you can appeal within 6 months at support@cinara.ai; appeals are reviewed by a person on the Cinara team.
We cooperate with law enforcement and regulators in response to lawful requests only, and report child sexual abuse material to the authorities.
Automated checks are not perfect. Today, a synthetic voice could read a consent statement. We are working on the next safeguards: detecting AI-generated recordings, comparing the voice reading the statement with the rest of the recording, and limiting how many statements one person can request. Trying to get around a consent check breaks our Acceptable Use Policy and closes the account.
Coming soon: a detector for audio made with Cinara, and content credentials on outputs that show they were AI-generated. Neither is available yet.
Report abuse
Email support@cinara.ai. We look at every report about voices used without consent, impersonation, public share links and copyright.
The owner of a link can turn it off at any time. We can disable links and voices that break our policies. Read the Voice Consent Policy.
Start with 10,000 free credits. No card needed.
Sign-up needs a confirmed email and a verified phone number.