Why AI Image Generators Struggle With Hands
AI generators evolve before our eyes at a scary pace, but they still have flaws. Spotting strange details in AI images is actually quite funny. That’s why Midjourney hands became a hot topic, a problem common in many engines.Let’s break down why hands challenge AI image generators so much. Their programmers are already fixing this meme-worthy issue, but it’s interesting to think about how artificial intelligence learns, not to mention what gets in its way.
Why AI-Generated Hands Made a Stir
Anyone using AI engines to create images may have noticed that hands rarely come out right, but the issue turned heads when a bunch of “photos” appeared on Twitter.
On closer inspection, the people’s weird hands gave them away as AI-generated images. The fact that this was Midjourney’s attempt at hands made the situation more interesting.
One of the best AI engines around could not tackle the intricacy of human hands, so the capabilities of Midjourney and its competitors were put to the test. True enough, even DALL-E is prone to unrealistic fingers and nails.
The hype was out of proportion, considering AI-generated hands have always been a problem, but the extra attention did prompt the release ofMidjourney v5 to improve on v4.
The new version made a point of enhancing hand design, a clear indication that AI engineers paid attention to the hilarious stir and decided to upgrade the software’s capabilities.
Other engines are slow to follow Midjourney’s example, sofixing AI art with Photoshopremains an invaluable skill. The main hurdle for programmers is how complicated it is to train artificial intelligence to draw convincing hands.
Why Do AI Image Generators Struggle With Hands?
AI engines use generative adversarial networks (GANs) or Stable Diffusion to produce images. Both technologies require extensive source materials, training, and processing power to create even the most basic artworks.
Since pre-existing images are central to an AI’s training, programmers have to feed their software thousands, if not millions, of pictures alongside prompts—repeating the process over and over again until the engine understands what a particular word refers to and how to represent that object.
But the source images an AI learns from are mainly 2D, where hands are depicted in a variety of positions. Whether straight or curled, showing five fingers or three.
At the end of the day, a machine doesn’t actually understand the concept of hands, and the pictures it learns from don’t always feature hands clearly or consistently enough. That’s why Midjourney hands can be so ugly: AI confusion.
As valid asElon Musk’s concerns about AI developmentmay be, some parts of the technology still have much to learn. And their obstacles go beyond insufficient examples of hands.
Other Reasons Why AI Image Generators Are Slow to Improve
Looking atMidjourney’s models, v5 offers advanced coherency between text prompts and produced images, as well as higher resolution and additional tools. But such achievements don’t come cheap.
Training an AI to do better with hands requires feeding it better images, especially in 3D. That means lots of time and manpower is spent on processes, from acquiring source materials to improving the coding and repeating the training until the AI gets it right.
Even then, the software can make mistakes in otherwise stunning works of art. Besides being a huge and complex job, it’s expensive. So, don’t expectfree AI text-to-image generatorsto step up to Midjourney’s caliber just yet.
Put simply, the problem with AI engines isn’t just about these computer programs’ inability to completely understand how human features like hands and feet look or work. It also comes down to what it costs, and the technology’s access to 3D imagery and machine learning techniques that can help generators get a more realistic grasp of the world around them.
AI Image Generators Won’t Struggle Forever
Hands are a tricky concept for artificial intelligence to wrap its binary head around, but solutions to the problem are already at work. Midjourney, DALL-E 2, and other platforms will eventually be able to keep quirky fingers at a minimum, if not eradicate them completely.
Advances in other AI fields ensure the technology is constantly evolving, and its developers always learning new ways to apply and improve it.
AI can recreate images from what you’re thinking. Is this an exciting development, or the start of a terrifying dystopia?
Some subscriptions are worth the recurring cost, but not these ones.
Revolutionize your driving experience with these game-changing CarPlay additions.
Obsidian finally feels complete.
Your iPhone forgets what you copy, but this shortcut makes it remember everything.
When your rival has to bail out your assistant.
