Member-only story
Artificial Intelligence, Design, and Visual Arts
How to fix prompts for AI-generated images of hands
And why AI art generators struggle with fingers and hands
[Scroll to the end for tips!] AI text-to-image generators seem magical. They allow creators to describe concepts and see the picture appear in seconds. But ask any AI artist what’s the hardest thing to generate, and we’ll all agree: we just want normal hands that don’t look like Halloween props.
Hands are one of the most complex shapes to generate. We’ve seen the fails when AI goes wrong and decides to give us the metaphorical finger! There’s a Discord channel dedicated to AI fails, and a Tumblr from one of my friends and colleagues. It’s safe to say we’re all morbidly fascinated.
Even with perfect 3D models and the most advanced AI systems, hands remain difficult to generate accurately. But the truth of the matter is — while there are technical limitations — good prompt engineering will go a long way to getting better results. Human error is part of the problem.
Let’s take a look at why it’s so hard for AI art generators to create realistic hands. And then I’ll give you my top-secret tips for how to prompt images of hands that will get a solid two thumbs up (and not seven freakish thumbs and an extra knuckle randomly sticking out of a hairy leg!)
Why are hands difficult for Artificial Intelligence?
1. Hands are amazingly complex body parts
Hand anatomy is complicated: Fingers come in all shapes and sizes and can be held in different ways. There are multiple articulations on each finger that need to be depicted correctly, or a hand will look unnatural. Even when the hands are in a “relaxed” position, there’s information to be drawn: wrinkles and folds on the knuckles, and shading on the palms etc.
Due to the complex geometry of hands, there is no standard set of lines or shapes that AI can recognize as a hand. In order to generate…
