The AI drawing hands issue has been going on for so long that it has now become a running joke. If you don’t know yet, spoiler alert: AI can’t draw human hands or feet properly. AIs that can convert words to images appear miraculous. They allow artists to explain ideas and watch a visual representation materialize in real-time. If you ask any AI artist what they find most challenging to create, they will always say the same thing: hands and feet.
This article is originally published on Jim Nightingale’s blog and was adapted to be posted here with the author’s permission.
Putting aside questions of ethics and logic, a major issue with AI artwork is its inability to depict hands and feet accurately. Despite the best efforts of the most sophisticated AI systems and the most realistic 3D models available, it is still challenging to generate accurate hands. However, while technical constraints exist, competent, prompt engineering will go a long way toward improving outcomes. But human error is also a part of the problem. So, can we fix it? Let’s take a closer look.
Why is AI drawing hands so badly?
Currently, artificial intelligence is infamously awful at generating some objects like hands. Although it can create unique artwork that sometimes rivals Evangelion, it still can’t draw human hands or feet.
There are several reasons for the problem:
- Hands are complex
- Hands are difficult to draw even with a pen
- Human perception
- AI is not the Superman
While there are still some debates about artificial intelligence-generated images, people are still looking for the best AI art generators. Will AI replace designers? The answer is still unclear, but first, it must overcome these problems.
DoNotPay AI lawyer is ready to give $1 million for any case in the US
Hands are complex
A variety of different-sized and shaped fingers allow for a wide variety of grips, making the hand’s anatomy quite intricate. Each finger has many joints that must be captured accurately for a hand to look natural. There is still detail to be drawn from “relaxed” hand poses, such as creases and folds in the knuckles, palm shading, and so on.
This is because the complicated geometry of hands means that there is no universal collection of lines or shapes that AI can use to identify a hand. AI must combine many various shapes and combinations to make convincing hands.
From finger length and width to metacarpals and wrist joints, the human hand has roughly 30 points of geometric variation. Did you know that the shape of a person’s hand may serve as a more reliable biometric identification than their facial features?
Hands are difficult to draw even with a pen
One of the most challenging aspects of traditional drawing is creating a convincing hand. They are among the most challenging things to represent because of the complexity of their geometry. Overcrowding of the hand prevents individual fingers from drawing distinct lines. Overlapping lines magnify even a minor error, making it stand out.
For this and other reasons, most cartoon hands only feature three digits and a thumb. Since we’re used to it, our brains don’t even register it; it’s like how we don’t even notice that The Simpsons’ house is yellow.
So, this issue with hands is not unusual or not only related to artificial intelligence. It’s a creative process. It’s funny and almost human that AI is suffering the same artistic growing pains that most artists do, such as learning to control their fingers and hands.
don't feel bad if you can't draw hands, even AI can't do it pic.twitter.com/89xEEQQvWv
— pikat (@pikatl) June 22, 2022
A further factor contributing to the unnatural appearance of AI hands is how humans perceive them.
The human hand and finger are unique. We have a natural, innate understanding of something about hands that we find highly interesting. Because of this, if there is an error, we will know about it immediately. The fact that hands are so intricate makes this problem even worse; if an AI creates a slightly incorrect shoulder, it doesn’t matter much because the whole inaccuracy is spread out. We might not even notice if a person’s shoulder was five percent smaller, but if their thumb, index, middle, ring, and little fingers were all a little smaller than average, that would be weird.
AI is not the Superman
Our expectations of machine learning are sometimes unrealistic. It is a relatively new area. We’ve been trying to perfect the art of sketching hands for thousands of years, and we still haven’t quite made it. A human artist may need several days or perhaps weeks to complete a realistic hand. When evaluating AI, we should keep this in mind and not have too high of expectations. Although current AI technology has the potential to provide stunning visuals, it is important to remember that it is still developing.
Obviously, it’s difficult for AI art generators to deal with all that information. Getting a realistic hand rendering that captures all these subtleties is tough. To make this situation more difficult, imagine creating an image of two people holding hands or a group of pals giving each other a bear hug; now multiply that by any number of hands you want to include. The whole picture will be thrown off if only one hand is off. Artificially intelligent hands can sometimes provide the wrong impression because of our overuse of them.
Anyway, there are some solutions to try to get better renders.
ChatGPT is at capacity right now; It is a really annoying error, but we know how to fix it
How to fix AI drawing hands?
AI drawing hands some of the biggest problems of AI art. Some strategies can help your AI art generator produce better outcomes, despite the complexity of hands. You need to know what AI can and can’t do so you can adjust your suggestions accordingly, such as:
- Hands position
- Create a scenario
- Get professional help
- Using reference photos
- Better prompts
Let’s take a closer look at the solutions.
Do you really need a hand? Try something new: To the extent possible, hide hands or have them be behind the characters’ backs or invisible.
Create a scenario
Artists with busy hands produce better work. As an illustration, consider a person with a cup of coffee in their hand. Training data involves defining a series of reference photos that depict fingers in a given configuration.
As obvious as it may be, don’t underestimate the impact of cropping. Carefully create the cut above the wrist. Avoid cutting near a joint (such as the wrist or elbow) unless you want your arms to appear like spaghetti.
With inpainting, you can remove a section of a computer AI-generated image and have AI restore it. In other words, it’s an excellent method for regenerating hands in isolation.
Survival #AI movie 🦕🦖made with Film Forge https://t.co/u1CH3CFCSy by @itsandrewgao. Script was made with #ChatGPT. I removed watermark from video to test Inpainting feature by @runwayml (sorry @itsandrewgao it was just for testing purposes 😉) #ML #ArtificialIntelligence pic.twitter.com/GMzswP1coJ
— Daniel Pikl (@danielpikl) January 13, 2023
Check out what is DALL-E 2 outpainting
Get professional help
Keep in mind that there are still artists whose career focuses on depicting hands in art. You should hire a freelancer if you need one. Even in the field of artificial intelligence, human creativity will always be necessary.
Using reference photos
It’s difficult to teach AI about the structure of a human hand without providing it with a model. People use MidJourney’s image-to-image generating capabilities to create consistent character or model images from the perspective of the topic. You can receive a more accurate hand by including a picture of a hand in your prompt.
Include tiny details like fingernails and knuckle creases as you describe the action and position. Define the hand’s contours, and use phrases like “curled”, “open”, or “knuckles up” for the position of a hand.
AI can’t draw hands, but it’s good at many things
Although AI drawing hands is terrifying, some AI tools are widely used and good at many things. Almost every day, a new tool, model, or feature pops up and changes our lives. We have already reviewed some of the best ones:
- Text-to-text AI tools
- Text-to-image AI tools
- Other AI tools
Do you want more? Check out the best free AI art generators.
Don’t be scared of AI jargon; we have created a detailed AI glossary for the most commonly used artificial intelligence terms and explain the basics of artificial intelligence as well as the risks and benefits of artificial intelligence.
Do you know there are also AI art robots? Take a closer look at Ai-Da.