The Limitations of AI Art: Why Hands Often Look Unnatural in Art

AI art generators have made significant strides in recent years, capable of producing stunning pieces of artwork that mimic the styles of famous artists such as Van Gogh, Monet, and Picasso. However, there is still one area where these AI algorithms struggle to accurately replicate, the human hand.

In a video produced by VOX, they offer three big reasons why AI art generators cannot generate hands correctly. One of the reasons why AI art generators struggle with hands is that hands are incredibly complex, with intricate details that are difficult for AI algorithms to comprehend. The shape and size of hands can vary significantly from person to person, and even small changes in the positioning of fingers or the curvature of the wrist can significantly alter the expression of the hand.

In the video, it's explained that while AI generators can have the ability to recognize the physical appearance of hands, they lack an understanding of the intricate movements and interactions that hands are capable of. They also do not have a comprehensive understanding of what hands truly represent. Essentially, AI generators identify hands based on the arrangement and combinations of pixels but, fail to grasp what hands are beyond their visual representation.

Check out the full video linked above for a detailed explanation.

Usman Dawood's picture

Usman Dawood is a professional architectural photographer based in the UK.

Log in or register to post comments
5 Comments

It is still very early days. Expect exponential improvement.

So interesting that in the 1973 movie Westworld the androids hands were not able to be accurate either.

Hands are fixed in the latest versions of ai art generators. Checkout midjourney 5 or stable diffusion.

This has nothing to do with complexity. Do you really think technology that can handle the complexities of faces can't manage hands? It's actually an analysis issue. AI replicates things by analysing millions of existing pictures, comparing them, and selecting something that corresponds to what it's analysing. So it understands that hands have more than two fingers, but it doesn't comprehend how many fingers a hand has, because it isn't always clear in pictures. It's for this reason that it also struggles with teeth and, to some extent, with hair. You also see it having problems with the number of limbs on certain animals that have prominent tails or trunks that confuse AI as to whether these are arms or legs.

As mentioned, THREE reasons were highlighted. I only discussed one briefly beause the content is the video and it wouldn't be fair to prevent the need to watch the video.

Hope that helps.