Generative AI images can sometimes look a bit…weird. This weird feeling we get has a name: the “uncanny valley.” Let’s dive into what this means and why it happens with AI images.

The uncanny valley is like a creepy ditch we fall into when something looks almost human or very familiar but is just off enough to make us uncomfortable. For example, have you ever seen a robot or a doll that looked almost like a real person, but something about it made you feel uneasy? That’s the uncanny valley at work. With AI-generated images, this happens when the pictures are close to looking real but have some oddities – like a person with an extra eye or a cat with oddly shaped legs. Or an octopus with entirely too many tentacles! These little mistakes can make us feel like something’s not right.

Generative AI is super smart, but it’s still learning from tons of pictures and trying to understand what we want. Sometimes, it mixes things up because it’s trying to copy and combine so many details at once. Imagine trying to remember every single detail of your friend’s face and then drawing it from memory. You might forget how exactly their smile looks or where their freckles are. That’s kind of what happens with generative AI; it tries its best but can still make mistakes.

Even though falling into the uncanny valley sounds spooky, it’s actually really important. It helps scientists and engineers figure out how to make these AI programs even better. By understanding what makes images fall into the uncanny valley, they can teach the AI to avoid these mistakes in the future. So, while those weird, almost-real pictures might give us the heebie-jeebies now, they’re helping us move towards creating even more amazing and realistic images down the line.

We can’t begin to tell you how many images we work on that simply do not work because of some issue or another, but other times, like with the picture above, the results are still quite interesting and beautiful in a way.