Stylised or Photoreal? Choosing the Look of AI-Generated People in Business Video

AI video tools now let a business choose how the people in its videos look: stylised, as in 3D animation, or photoreal, close to filmed footage. This paper asks which look fits which purpose. We review four bodies of evidence. Research on the uncanny valley gives limited support for a general dip in liking as characters become more human; the better supported risk is a mismatch between realistic and unrealistic features. Advertising research shows that non-human spokes-characters can match or beat human spokespeople and build trust. Perception research shows that AI-generated faces can be indistinguishable from real ones and are sometimes judged more trustworthy, which turns photorealism into a question of deception; the EU AI Act responds with disclosure duties for deep fakes. Finally, experiments on AI disclosure in advertising find that revealing AI generation often, though not always, lowers trust and ad evaluations. From this evidence we derive a decision framework that matches the look to the purpose, and we argue that AI-generated people should never stand in for real customers or testimonials. We illustrate the framework with GetReelCom, which offers an animated and a realistic mode, described from its public documentation. No study yet compares stylised and photoreal AI characters in business video directly; we propose one. Competing interests: the author is a co-founder of GetReelCom, the application used as the illustrative case. This paper is not legal advice. About the author: Amber Kesselaer is a growth consultant and an expert in marketing and AI video generation.