The Uncanny Valley and Character Design

Attention, Internet: I have a new article on the psychology of the uncanny valley up on You know what the uncanny valley is, right? It’s that theory originally from the field of robotics that says if you stick a couple arms and googly eyes on a trash can it looks cute, but if you don’t get facial animations or movement right on an otherwise realistic looking android it looks creepy as hell.

Nathan Drake and the traveler from Journey represent the two high points on either side of the uncanny valley.

Nathan Drake and the traveler from Journey represent the two high points on either side of the uncanny valley.

This has implications for the design of characters in video games, and the uncanny valley is sometimes cited as the reason why opting for more stylized character designs is a better choice –especially if you don’t have the budget and expertise to do tons of motion capturing and super high resolution textures. In the last few years some psychologists have done research on the underlying causes of the uncanny valley, and in my article I look at some of them and see what implications they have to say about character design in games.

I gave our friend Bobo the Quote Monkey a map of the uncanny valley and sent him off for a quote from the article. He came back looking a little freaked out and clutching this:

It shouldn’t be surprising that faces are one of the most important things determining whether or not a video game character will live in the uncanny valley.

One study by Karl MacDorman, Robert Green, Chin-Chang Ho, and Clinton Koch published in the journal of Computers in Human Behavior suggests this is true and provides some specific guidelines for those character creation tools we love to see in RPGs. In one of their studies, the researchers took a realistic 3D image of a human face based on an actual person. They then created eighteen versions of that face by adjusting texture photorealism (ranging from “photorealistic” to “line drawing”) and level of detail (think number of polygons). Study participants were then shown the 18 faces and asked to adjust sliders for eye separation and face height until the face looked “the best.”

The result? For more realistic faces with photorealistic textures and more polygons, participants pursued the “best” face by tweaking the eye separation and face height until they were pretty darn close to the actual, real face the images were based on. But for less realistic faces with lower polygons and less detailed textures, the ranges of acceptable eye separations and face heights were much larger. In a follow-up experiment the researchers did the same thing, except they asked the participants to adjust the sliders to produce “the most eerie” face instead of the best one. Again, when faces were more realistic looking, it didn’t take much tweaking to make them look creepy, but when the faces were more stylized and less detailed, a wider amount of facial distortion was acceptable before things looked eerie.

Read the whole article here. If you like it, please comment or share it on your social media outlet of choice.

7 thoughts on “The Uncanny Valley and Character Design

  1. Have you seen the first episode of the second season of Charlie Brooker’s Black Mirror series? Highly recommended viewing.

    It deals with the uncanny valley in a way I haven’t seen before. The argument is that if you produce a robot that looks, moves, smells, feels and sounds like a human being — hell, even if it can easily pass the turing test — there is still plenty of room for dissonance.

    … Which is what bothers me with ‘realistic’ games. It’s not just that the higher fidelity your graphical representations are the more you can pick up internal inconsistencies, it also lampshades shortcomings in other components; most of all behavior. AI in nearly all games is a joke and a misnomer.

    The result is that folks like Birgirpall ( can utterly, beautifully destroy your game.

    It sends a message: why even try? I mean, realistic graphics are an interesting goal in and of itself, but perhaps not when you’re trying to craft an internally consistent world.

    • Huh, I’ll have to check that out.

      There’s probably more to the phenomenon when you add interactivity to the mix, and I’d love to see researchers address that. Usually, at no point are you left wondering if another character is controlled by a human, because they do things no NPC would ever do. That said, though, I remember when I first played Journey and encountered another player. He was just shuffling around a limited area and at first I wasn’t sure if he was being played by another human or not.

      • Good point regarding Journey! It’s much harder to tell the difference in Journey because of limited actions available to a Journey ‘puppet’, and its playful nature results in a less strict understanding of what is ‘wrong’ behavior. The playing field is more level (less parameters for the AI to worry about), so you could very well expect an AI to be driving the other character you encounter.

        It’s easier for the AI to pass an at-a-glance turing test in Journey than in, say, Crysis. Just like researchers have found this to be true for audiovisual representations, I would also argue it’s easier for a stylized character to *behave* in a believable way than for a ‘realistic’ character.

        So yeah, if I were to do game AI experiments I would probably create a world inhabited by stylized amorphous alien creatures since there’s no existing expectations for interacting with them. The polar opposite of something like Heavy Rain. 🙂

        • I’m not sure. If I remember correctly, it looked like they were AFK at first, then they got up, rotated around (like they were getting their bearings), chirped at me, then ran off. I just assumed they were human at that point.

          • I wonder if That Game Company did any research/documentation on how players perceived other other players, and whether they identified specific common tells that gave it away for people.

            I’m now also wishing Journey had an API so people could do social experiments with it, hah. Like, if you could codify some of these tells and get an AI to perform them, could you get an AI to pass the turing test within a single playthrough of the game.

  2. Pingback: Dev Links: Spinning Plates | The Indie Game Magazine - Indie Game Reviews, Previews, News & Downloads

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.