Google Gemini Flash TTS can clone a speaking voice from roughly 30 seconds of audio, and that is now a live risk for brand voiceovers, not a future demo.
The feature sits inside Google’s broader Gemini generative media stack. You feed a short clip, then ask the model to read new lines in that voice. For marketing and PR teams, the upside is fast localised drafts. The downside is consent, disclosure, and deepfake policy gaps that most brand books still treat as optional.
If your agency already scripts AI drafts in Gemini chat, this lands next to the tools your writers use every day. That makes the ethics brief urgent, not academic.
What changed for creators and brands
Voice cloning used to need a studio pack and a signed talent deal. A thirty-second sample lowers the bar to a WhatsApp voice note. That helps small creators mock talent reads before a shoot. It also lets someone paste a CEO keynote into a spoof ad without asking.
PR leads should treat this like photo deepfakes did two years ago. Update the client crisis sheet now. Name who can approve a cloned voice, how talent consent is stored, and when a post must carry an AI label.
For MENA brands, add language rules. A cloned Arabic or Gulf-accent voice can travel faster than legal review. Keep a one-page checklist in the creative room so juniors do not ship first and apologise later.
How teams should brief clients this week
Lead with use cases, not fear. Say: we can speed scratch reads and A/B tone tests. Then draw the red line. Paid media, founder voice, and celebrity partnerships need written consent and a watermark plan.
Ask legal for two templates: a talent addendum for AI voice doubles, and a creator brief that bans training on client audio without a signature. Soft-source any vendor claim about “safe” cloning. Policies change faster than product pages.
If a rival brand already posts AI voiceovers, do not match them blind. Match their disclosure standard, then raise yours. Trust is the product you are really protecting.
FAQ
Does Gemini Flash TTS need a full studio sample to clone a voice?
Public positioning points to a short clip around thirty seconds. Treat that as a capability headline, not a licence to skip consent.
Should brands disclose AI voice clones in ads?
Yes when the audience could think a real person spoke the line. Put disclosure in the creative brief, not only in the footer after launch.
Google just made voice cloning a Tuesday task. Your ethics memo should land before the first client asks for a “quick clone.”
