Fumbling with Fingers: The Difficulty of Generating Realistic Hands in AI Imaging

AI imaging software has made a lot of progress in making realistic images of people, but some body parts, like the hands and feet, are still hard. Human artists have the same issues with these appendages because they contain a lot of organic parts that shift and change in perspective.

Drawing Hands and Feet

Hands and feet are highly complex structures with many bones, joints, sinews, and muscles. This makes them difficult to draw, model, and animate.

Hand and foot movements and gestures are also hard to copy because they can show a wide range of feelings and intentions. You can create a library of hands and feet references (I’ve had them for over 40 years) and still not scratch the surface of all the variables involved in creating hands and feet.

Sculpting Hands

Sculpting these appendages is even tougher because now you’re working in three dimensions. There are cheats that can be used when drawing and painting, but it’s harder to cheat your way to a good sculpture.

The best option for anyone who gets into art or image generation is to learn the basics of anatomy and design. If you’re doing this for fun, it’s just going to make your work more interesting than the million other Waifu images out there.

Human Artists are Still Better

There’s been a lot of chatter about AI imaging software stealing other people’s artwork on the Internet. The fact that hands and feet still look terrible after the various AI models out there have viewed billions of images is probably the best indication that this is malarkey.

Professional artists who draw figures understand and practice the basics of drawing, painting, or sculpting hands and feet for a long time before mastering those skills. At this point, they can beat any AI system at creating eyes and human appendages. It shows in the samples below. That will change a more artists train their AI assistants to draw in their styles.

Data Priorieties

AI systems use information from many different places to make images. Data can come from many sources, such as images, 3D models, videos, and text that already exist. AI systems use this information to teach machines how to spot patterns and make new images.

This information is used to make models that learn how to make realistic pictures. These are based on the information the AI apps are given. AI systems can also use data to make new images from scratch or to improve or refine images that you already have.

However, the folks who created these apps didn’t really see the need at that time to get the systems to refine certain organic details. Things like eyes, fingers, and toes take a lot more time and resources to perfect.

Training data for hands and feet is often limited, making it difficult for AI models to learn the intricacies of these body parts. Collecting data is also hard because it requires high-quality pictures or videos of hands and feet in different positions and lighting.

How Artists Learn to Draw Hands

Imagine a person getting to view billions of bits of information over many years. You now ask that person to recall a picture they saw. Tell them to create it. Give them some hints as to what was in the picture you have in mind. Go ahead and give it a try with a piece of paper and a pencil.

We start with the most basic shapes in the most basic positions.

Then we build on that shape with a bit of refinement.

Through training and observation, we keep refining our skills until we have something that people recognize as hands, eyes, feet, or whatever our project calls for.

To improve their drawing skills, many artists also practice sketching hands from life, either from their own or from a model. They may use different methods, such as gesture drawing, which involves quickly sketching the hand’s overall shape to show its movement, or block-in drawing, which involves breaking the hand down into basic shapes and forms.

Many artists also use reference materials, like photos or drawings of hands, to learn about different poses and lighting. As they practice and get better, they learn to notice and understand the small details and subtleties of the hand. This lets them make drawings that look more real and have more meaning.

AI and Commercial Art

Commercial art and photography are already benefiting from AI applications. Artists and photographers have been using digital tools to create better efficiency when it comes to production. The name of the game is getting paid and moving on to the next job. Any tool that helps speed up the process, make it easier, and offer fewer headaches is often welcome.

Whether drawing, painting, sculpting, or creating 3D models, I rarely start from scratch on things like heads, hands, eyes, or feet. There’s no need to establish a base for a project when working quickly.

Current Solutions

That’s where swipe files, kitbashing, and AI imaging apps come in handy. Professional content creators can draw from a variety of tools and resources to get the job done in the least amount of time with the best results possible. So, what are some existing ways to fix the issue with AI software and its hand problems?

Hide Your Hands and Feet

The fact is, until the software gets a bit better, you can create prompts that avoid poses where hands and feet are visible. For example, a prompt that states something like, “Imagine a young man with his hands in his pockets looking at a beautiful butterfly.”

Use Image Editing Software

I’m a control freak, so I rarely use AI imaging software on its own when working on projects. This is also true when rendering images from 3D software on its own when working on projects. This is also the way I work when rendering images from 3D software.

There’s no need to settle for something less than what you want when the tools are available to get things right. There are free and low-cost alternatives to Photoshop that can work for post production or in concert with your AI renders.

Here is a list of software that you can use for post production or just image editing:

1. GIMP (GNU Image Manipulation Program) — Free and open-source
2. Inkscape — Free and open-source
3. Paint.NET — Free for personal and educational use
4. Seashore — Free and open-source for Mac users
5. Pixlr Editor — Free online image editor
6. Canva — Free and low-cost plans available
7. Fotor — Free and low-cost plans available
8. BeFunky — Free and low-cost plans available
9. Photoscape — Free for personal use
10. Krita — Free and open-source

Researchers have come up with different ways to deal with these problems, such as using 3D models, multi-view images, and data from motion capture. It’s going to take a while for those improvements to reach acceptable levels of quality for the pros to appreciate.

Some researchers also use both data-driven and rule-based methods together to make hands and feet look more real. I predict that the improvements will be up and running by the end of 2023.

Conclusion

AI imaging software has made a lot of progress in making realistic images of people, but it is still hard to do because hands and feet are complicated and there isn’t a lot of training data. I’m still of the opinion that non-artists should spend some time learning the basics when they commit to this new hobby.

It’s going to take a lot of resources to address these issues. More artists will continue to come forward to help in the process of improving the realistic appearance of AI-generated hands and feet. I’m currently looking to work with others to create models that improve hands and feet for less hideous renders.

I often write about ways that content creators can generate income through side hustles or build their businesses. I cover marketing, content creation, and how to write for a living.

Read more of my work here on Medium or head over to my blog at https://contentmarketingmagic.co/blog. I also own a SaaS company that produces content templates, swipe files, and article samples for small business owners who want to write but need that extra push. Sign up for a free pass at https://contentmarketingmagic.co