AI Video Explainers and the Uncanny Valley
Tools like Animaker, Vyond, and PowToon are popular in K‑12 education and corporate training alike for their ability to quickly create talking head videos out of a simple chunk of text. But are these easy video generators really the best option? This post is a case study of a recent project I completed where stakeholders asked for an animation in this popular style, but after seeing the available options, they ended up preferring something much more low‐tech.
Stakeholders Request an Animated Talking Head Video
Recently I was approached by stakeholders to create an animated explainer video based on some recorded Zoom meeting footage they had captured. They wanted me to keep the speaker’s audio but replace the video with an AI‐generated animated “talking head” avatar.
Presenting Stakeholders with Options
Animaker
I first did what they asked – I fed the source audio into Animaker and produced a just‐what‐you‐expect talking cartoon person set in a modern cartoon office, mouthing words along to some jaunty ukulele strum music. It was fine. You’ve seen them before. Whatever.
Kaiber.ai
I wanted to also present the stakeholders with something they’d never seen before. I fed free copyleft‐licensed video footage from Pexels.com into Kaiber.ai to produce stylized versions of the footage, then edited that together with real human speakers in a conventional video editing suite to create something more compelling and eye‐catching. I produced this proof‐of‐concept using that process in only a couple days:
Simple Homemade Motion Graphics Video
As a third option, I simply designed a nice static screen for each presenter and left it up on screen as they spoke, sometimes with minimal bullet points or subtle stock video in the background to emphasize key points. I modeled this approach after the way newscasts have traditionally shown speakers when live video is unavailable.
In the end, the stakeholders preferred the simpler, non‐animated version.
I think this is instructive, so let’s reflect for a moment…
Navigating the Uncanny Valley
The stakeholders were sensitive to the “uncanny valley” in the auto‐generated options – the dead eyes, the lips moving that don’t form recognizable words, the odd voice inflections. As close as these options come to realistic, the imperfections take over my attention when I should be really paying attention to the content.
I recall learning in my psychology classes that the brain has a disproportionate number of neurons dedicated to picking up details of the human face. This makes us incredibly good at recognizing people and reading emotional cues on faces. It’s part of the reason why we see faces in inanimate objects like clouds and ink blots – being able to perceive faces (whether friend or predator) has been critical to our survival over tens of thousands of years. It also makes any unusual characteristics of the human face stand out to us, producing strong negative reactions, and contributing to the “uncanny valley” effect.
If you decide to use any of these AI video explainer generators, you may need to consciously work against the brain’s refusal to accept their not‐quite‐lifelike appearance.
Stepping Back from the Uncanny Valley
In Kevin Cheng’s book “See What I Mean”, where he advocates for using less‐detailed representations of people wherever possible. The brain (as we see above) is very good at filling in the details of a simple line drawing of a face. As the representation gets more detailed, it also takes up more of your brain’s attention that we ideally want to be directing to the content being learned.
For this reason, I think it’s a best practice to default to less‐distracting representations of speakers, even low‐tech static images, rather than AI‐generated ones. I would even say that low‐quality video of a real person is better than high‐quality AI video of a dead‐eyed talking robot head.