Why AI Art Struggles With Hands | The Physics Behind AI

wow isn't everything just stunning and perfect in this AI generated video hold on a second what's that let's watch that [Music] again are those fake hands AI is amazing until you really focus on those pictures and videos and notice some odd mistakes in the [Music] hands it's weird to think that after getting everything else right like the face background and overall appearance AI often struggles with hands why is it always the hands and those Strang looking fingers now look at this photo I asked Del to create a picture of two boys shaking hands and see what it came up with then I asked Chad GPT to describe the image to me it went into detail about the boys wearing school uniforms and the Blurred background but what's absurd is that it missed mentioning the extra fingers I mean just count them why not ask directly and get the answer when I ask directly CH p 4 explained that it only appears like there are more fingers but upon closer inspection it's an optical illusion due to the angle and there are actually 10 fingers in total so basically we humans are just fools at this point I feel like why did I even buy chat gp4 so what's the real reason behind it why AI struggles with hands the reason hands are a real challenge for AI to get right because they're the Pinnacle of complexity when it comes to human movement and expression and think about it our hands can twist grip and gesture in countless ways that we instinctively recognize but AI has to learn from scratch and it's not just about movement it's about the subtlety of those movements which are as varied as the people making them the way your fingers fan Out To Catch A Falling Leaf is poetry and motion to a human Observer but to an AI it's a puzzle with a million Solutions another possible reason is that the Learning Materials AI uses to recognize and replicate images of hands aren't always ideal in many photos the hands might be partially hidden or not the clear Central subject of the image they could be in motion which often results in a blurry picture or tucked away in the background as a consequence AI doesn't get the comprehensive detailed pictures it needs to fully understand and accurately recreate hands in different positions and activities how many fingers do you see right now like two or three so it's like it doesn't know there's five cuz sometimes there's two sometimes there's three sometimes four sometimes five adding to the complexity has hands are just so cuz both your hands are going straight up right now well hand likee now take a moment to really look at your hand it might seem normal because you see it every day but if you think about it it's kind of weird you've got five fingers sticking out of your wrist and your hands are full of interesting details this complexity is why AI sometimes has trouble drawing hands correctly difficult very difficult we also can't Overlook just how good we are at spotting when something's off with the hands so even the slightest mistake that AI makes stands out to us now look at this picture when I ask chat GPT to add five pimples on her face he literally adds so many pimples on her face well this is not what I asked for but it doesn't matter that much because after all it's okay if it's not five we can still use it however if we run into the same issue with the fingers that's something we can't ignore right what's wrong with you what's wrong with you one user said as someone who likes drawing it's not just the AI hands are a nightmare and another person said as a drawing artist myself I can tell that hands are so difficult to draw I rarely draw them for that reason these comments suggest that both a I and human artists are in the same boat when it comes to the complexities involved in drawing hands making it a universal artistic challenge however some people also shared their views that it doesn't happen anymore and that AI gives perfectly smart hands one person said when was the last time you tried generating art they can do hands just fine now so many people think that AI can draw hands just fine now without problems but the images I showed in the video are ones I made very recently and you can clearly see that they still have mistakes it's important to remember that it doesn't happen every time but when the hand positions are complicated the AI tends to mess up in the same ways if we expand our view Beyond still images to videos we'll notice similar issues take the videos generated by Sora for example this AI system is at the Forefront of video generation technology even though Sora is very Advanced it still struggling with getting complex movements right especially those seen in the latest videos from February 15th 2024 this shows there's still room for improvement Sora is not just any tool it's meant to be one of the best out there but this issue reminds us that even the best systems can still get better especially in mimicking how people move despite these challenges Sora is still ahead of other AI Tools in making videos and images what sets Sora apart is its use of physics to make its videos feel more real unlike other tools that just aim to make things look good based on what they already know Sora takes an extra step it doesn't just create images or videos it makes them interact in a realistic way with their surroundings this focus on physics could really change the game in content creation capturing the full range of human movement and detail has been a longstanding challenge not just for artists but now for AI as well while there's been a lot of progress it's clear that AI still has a long way to go it's getting better but it sometimes still gets things wrong like placing Parts in odd positions or not getting the details quite right these errors highlight that a AI needs to improve its grasp of human anatomy and movement as AI technology improves we expect it to get better at capturing the complexity of movement and detail understanding these challenges helps us appreciate the complexity of what's being attempted and the remarkable achievements of human creativity throughout history well what do you think of it share your thoughts in the comment section below and check out the videos on your screen for more interesting and AI related content [Music] [Music] [Music] [Music]

Channel: Technomics

Views: 790

Keywords: ai, artificial intelligence, midjourney, pika labs, pika 1.0, runway ml, ai art, text to video, prompting, prompt engineering, sora, ai image generator, ai video generator, ai image generation, ai video generation, physics engine, openai, conputer graphics, cgi, ai art vs hands, why cant ai draw hands, stable diffusion, ai art generator, ai art hands, machine learning, generative ai, generative ai in creative industries

Id: uCTNk9HiYe4

Length: 8min 1sec (481 seconds)

Published: Mon Apr 15 2024