I just woke up and really surprised to find this video got 3 digit views in a day! Thank you so much lads! Edit: It's already 7k 😭😭😭 this is an achievement for me, thank you! Edit 2: ALMOST 70K AAAAAAAAA 😭😭
as a human and very much not a robot, i am unable to relate to such urges. i only perform activities such as putting apples on thi- i mean eating apples with my human mouth
Image AI tends to be trained by cutting it loose on the Internet, as it takes a *lot* of input to grasp concepts like basic human anatomy. As a result, the models are kind of predisposed to believing cats are everywhere and women are frequently naked. They learn from our example. So, y'know, Skynet is going to be way more interesting than we expected.
the apples, the cats, the random near nudity, the ABSURD NATURE of how chaotic some frames are - as scary as ai art is to me as an independent artist, this is a masterpiece to me lmao
The AI really fixated on the word "apple", it makes the final result quite beautiful actually, this constantly shifting thing with the two recurrent elements being apples and girls... and the occasional apearance of a cat for some reason.
An educated guess: The AI model fixated on drawing an apple, but at the same time the term "bad apple" has been catalogued as manga/anime related (see how every character is drawn in manga style despite the prompt not asking for it). And what's the biggest collection of "free" manga style pictures that the model could learn the style from? Porn doujins...
@@zpacula Maybe, could also be, because AI image generators that make images fit to a prompt basically have to change the image with a few pixels, then check if it makes it fit to the prompt more or less with a recognition AI, change the pixels back if it doesn't fit better and repeat that a bunch of times, but not as often as regular images that are being generated from noise, because it still needs to look like the original image you are trying to change. This way it has a tendency to not want to change big white blobs from the middle, because it's like it makes the image make less sense for it to have random pixels in something that looks like it should be fully white. So often smaller shapes get more colors and it ends up with something that almost always uses similar colors for the bigger objects. It's possible the white colors ended up being used as skin, because that is closer than colorful clothing to the original pixels and "has an easy path to reach" by morphing the image to fit more and more to the prompt. Well, this also causes it to not move away too quickly from the original shapes and styles (unless you let it change the images for longer), meaning it might have recognized that something that looks like it is closer to a manga-style probably doesn't have random other styles in the same image, so the pixels shifting towards that might get removed again. Whatever, that's just what I think is going on.
Don't like to ruin joke, but it's probably discomfort from unnaturallity, same as fractals do to some people. This looks between natural and highly impossible in real life
I love how the ai is like “did it look like this?” For literally half a second and then change the design completely and is like “no this still isn’t it” and it does that the whole video
AI can't stay consistent, unlike sentient artists If AI used it's own images for reference (just like regular artists) to stay consistent, it wouldn't be random chaos But it's still not sentient, so it doesn't really knows the context
Honestly, i liked the part with Suika, the way the AI interpreted the sake drop as an apple was nice, the way the apple turns into a bottle of some sort that Suika drinks from was also awesome!
Man the AI really does love processing certain... features xD Even with the struggles you went through, this is Beautiful work from the AI. Any faster and it might not be able to to capture the characters properly.
I think it thought that the black line between the character's dress and the ribbon on her back was an asscrack. For some reason in many parts of the video the AI *really* wants to make it NSFW.
8:30 Ok, can we just appreciate how smooth this part looks? Like, the apple falls straight into the line and then a larger apple opens like a ripple effect. I don't know I just find this part so cool
Now I'm just imagining putting these frames through one of those interpolation algorithms (I think one of them is a GAN?) to bump it back up to the original FPS, with shenanigans sure to follow.
I love the way the AI incorporates the apple in some places, like the bottle, the doll, the ying and yang symbol, some of the clothes and hair, and a few others.
9:20 I like how the AI is so dedicated to trying to make risqué anime girls that the scythe briefly turns into her arms. Also it kind of unintentionally created some jiggle at 9:22 lmao
that one Frame at 8:54 / 8:55 honestly is creepy The " Bottle " turns into a girl but the girl drinking is still there so u just see a Girl crawling out of another Girl's Face . like that's just disturbing I can just visualize it in color and my God that's some Fk-ed up Shet .. xD
@@epicosexio I needed someone to share my suffering so I brought it up , and I wanted to point out something odd nobody mentioned cuz everyone else was already talking about the Marisa cat xD
@@coxinhalover that wasnt exactly true btw, I just watched it once at 0.25x speed before reading your comment, and only saw shit that couldn't be a wallpaper, which is why I made that reply when I saw your comment Im rewatching at 0.25x speed again, and already found one that is not that bad
As much as I despise what this kind of AI is doing to art, this is such an aesthetically cool video that you ended up with. The mish mash of elements and it attempting to match whatever frame it was imitating produced some really uniquely beautiful and surreal images
I agree; I think the difference lies between when someone tries to ai generate art just to show people bc it looks cool vs they try to claim that they "drew" it or whatever or are getting monetary gain. I actually think stuff like this is super cool!
@@ambyandony The eventual end goal of AI art groups in the current era is to replace all artists, since an AI license is a billion times cheaper than hiring an army of artists for your projects.
Luckily people don't have to worry too much about ai taking over anytime soon. The majority of its "art", is just taking other people's art and changing details about it by taking more people's art. It's not exactly creating something from nothing.
@@Hellfox1134ogPrivate That's a problem too, though, bc those people will never get credit for their work that's getting largely just completely copied :(
5:20 I am pretty sure the "AI" did that because if a character has clothing it's a pretty much decorated with colors and lines or tiles or something. The og frames of bad apple are flat and there is no decoration at all. So it is more close to nude character's skin (because those are kind of the same color too). Idk where you got the data to train on but it probably contains a lot of such pictures too.
Love that it sticks with the idea of interpreting every vague circle as an apple, excepts at 8:16 where the ai decides it wants to do googly eyes instead
I had a dream last week where it was a trend to take popular video and take every frame through AI with a horror prompt. The results were incredibly disturbing.
You truly know that AI has reached the peak of evolution and technology when heads spin, castles turn into bottles and apples, and a apple is in each frame lmao.
What Touhou fans think Bad Apple is: *Complex lore music video that brings out the most about some Touhou characters* What normal people (or the AI) think Bad Apple is: *APPLES AND HOT ANIME GIRLS*
I know many people have said that before me but I just love how the AI remembered that the title is about an apple and tried to add it basically everywhere
found this video within the youtube algorithm, i will happily watch this channel grow. your efforts to even make this at all are *impressive* and i definitely don't have that kind of motivation and patience.
nah that "thing" will a 5/6 of the frames by itself and let me just say it even with just %25 of the frames added doesn't look good if you look a little closely and when you see it once you can't unsee it so just no
I loved how on parts like Kaguya (the second one reaching towards the moon) and Yuuka (the one with all the flower petals flying around her) the AI understood the assignment and it translated very well. Probably because it was a faster movement than normal in the video. This was very cool to see! Thank you for your hard work!
A positive side to the 5 FPS, it makes it possible to process the separate frames. If it was going at 30 FPS and just as chaotic, it would be less good. Random cats need more than subconscious flicker time, and I'm glad they got more.