Influencer Cloning: An Ethics-Adjacent Workflow

1. Sourcing the Voice

We start with roughly 14 hours of podcast appearances. Most A-tier influencers appear on enough podcasts to provide ample training data, and unlike licensed TV they have not, historically, registered any objection to its use. Our voice model can, after a weekend of fine-tuning, deliver any line of brand copy in a rendition indistinguishable from the original, including the small mid-sentence sigh that their audience interprets as authenticity.

2. Synthesising the Face

For video we use a single 4-minute frame of footage from a YouTube cooking segment. From this we generate 14 distinct synthetic personas: same face, slightly different cheekbones, slightly different hair, slightly different body language. We license these out as separate creators to non-overlapping clients in non-overlapping verticals. The originals have not, to our knowledge, noticed.

3. C&D as Marketing Collateral

Once or twice a year a creator will discover the doppelgänger and serve a cease-and-desist. We have built our PR strategy around this. The C&D is itself a viral asset; the original creator’s indignation drives free attention to the synthetic creator’s product placements, which often outperform the originals on conversion in the days following the dispute.

We now budget for an expected three C&Ds per quarter as a media line-item. They have, on a per-impression basis, outperformed our paid TikTok spend by 11x.

4. The Successor Question

We have begun asking ourselves what happens when the original creator stops being a creator — retires, ages out, or, in one regrettable case, dies. Our current view is that the synthetic version simply continues. We do not anticipate audience backlash, in part because, by then, most of the engagement on the synthetic version is itself synthetic.