Bad Apple, but it's rendered and colorized with ControlNet [Stable Diffusion Deforum test]
ฝัง
- เผยแพร่เมื่อ 13 ต.ค. 2024
- Obviously, based on the famous clip • Bad Apple!! - Full Ver... for the Touhou song 'Bad Apple!!'
Used Anything v4.5. The initial rendering was accomplished with canny edge detection and 'scribble' ControlNet model, then the resulting grayscale video was put again through CN, but now with 'hed' preprocessing and hed model to get the coloration.
Deforum settings files: render - gist.github.co..., coloration - gist.github.co.... The black and white canny edges + scribble-mode mid-stage rendered video . You can check the prompts keyframing file to see if I got the characters names right. Despite my best efforts, some of the characters appearances were off, I think it was either the character names were incorrectly spelled (I was referring to the Touhou wiki) or they had been underrepresented in the dataset or there was overlap between their naming and some other franchises. There is also the issue of random unrelated characters appearing when the prompt contains no characters and instead is purely abstract, I guess it relates to this specific model retraining.
The pure rendering process took around 12 hours, but it actually stretched to around two days because of my computer crashing multiple times amidst the process or the animation being completely off. To make it more smooth and accessible, I reduced the frame count threefold at first and then interpolated the video back using another free and opensource RIFE AI-model. Then upscaled the video x2 with realesr-animevideov3.
This video was made for the purpose of exploration and showing off the new amazing ControlNet feature in Stable Diffusion AI-art system using the frontend of Deforum and AUTOMATIC1111's webui. Any similarities with Apple logo are purely coincidental and related to the model's bias.
Feel free to ask me any questions on the process.
This honestly feels like I’m in a dream
Edit: how the hell did I get so many likes, thanks y’all
It kinda makes you trip
This feels exactly like a dream
Ikr
same, this made me dissociate a bit ngl
RIGHT?!
2:20
i like how the AI actually turned this into a concert
it knows what's up
Ai is cool i guess
But the instruments still make no sense. (With the exception of the keyboard.)
The fact it's actually a performance makes this better.
Bro, use a character lycoris or Lora. This is dated compared to modern technique.
these transitions are smoother than my life
I'm scared for your life then
Zamn
you’re kidding right
Although its a nice example of what AI is capable. This is just a discrase to the original animation
@@spicypills666 no :)
I love how you can tell it knows who most characters are (Reimu, Marisa, Yukari, Mokou, and Youmu are good examples of this) but it has no idea who Eiki, Eirin, and Yuuka are.
Edit: ...oh...my...god... 1.1K LIKES???? I'M FAMOUS! This is the most likes I've ever gotten on a comment, thank you all so much!
Edit 2: Look, Mom, I'm even more famous! 2.2k likes! I hate to be the person who's all like "OMG TYSM FOR THE LIKES GUYS" but this is genuinely insane!
Yeah, and that's weird since that's Anything 4.5 and was supposed to be trained on lots of Danbooru and Pixiv images 🤔 I just guess they were underrepresented in the dataset (and I don't know of which it consisted of exactly, since anime models rarely disclose their sources). Or there are some prompt problems, although I checked the naming twice on Touhou wiki
it put tenshi on a step stool 💀💀
@@marisakirisame1st ahahaha how embarrassing lol. Tenshi on a step stool can’t be good for her image xD
True
@@marisakirisame1st smol tenshi XD
4:23 to 4:35 is crazy
It's just so smooth and all
I cant even describe it in words lol
I believe its hand stitched images
Honestly all the teleporting around and random character changes seems kinda on brand for touhou
Lol
5:06
I love how AI in the end tries to see complex stuff and can't see basic symbol
2:39 i love how koishi appears out of nowhere around this part for a brief moment, it's so in character fr
"Fine I'll do it myself" - Koishi
"Me not in Bad Apple? Fine! I'll put myself in Bad Apple!"
This whole video is Koishi mindgames
flandre is in it too for some reason
Well, Koishi is Koishi tho
3:00 To the eye is hella cool oh my gosh.
Holy shit, for an AI that doesn't remember the past frames the section from 1:50 - 2:34 feels very continuous and actually flows well
...
I dunno why, but it felt like a dream. I mean... Like a typical dream where something happens and you don't understand what is it exactly. An untoldable chaos without a main detail of the story.
Ok, I just watched it and felt like I was sleeping😂 hope you got it...
It's true and it's because AI can't generate something concrete.
Same feeling man I feel like i am falling asleep
I just woke up from a nap so it’s even more trippy
so true, i feel like i just woke up
same
Love how the characters occasionally got out of their own silhouette
and the fingers intersecting at 4:56 is quite shocking (the original animation could have been interpreted as fingertips touching)
It was probably prompted as "handholding"
So lewd!
What do you mean, in the original animation you can clearly see them cross fingers with each other
@@dection8948 yes, however most AIs that generate images often obscure original images with noise then generate new ones by carving the noise. This can make it hard for the AI to see small details to copy
hands are the easiest way to spot if something is AI or not
Nice, but I think one problem Is that he can't really Remember the previous frame(the Castle for example) and so there isn't a lot of continuity from One frame to the next. Still Very very good! I liked it
Seems to be a problem with ai in general, which is why making animation frames is so difficult. But don't worry, it will be improved upon soon enough, the programmers have been going at improving AI art like crazy, it's almost scary
i love how you personified the ai with "he" pronoun :)
@@cryptikkcries a lot of languages use “he” as a default pronoun for every word, gendered languages exist
@@KopieOG buddy i know, i speak two of them
@@cryptikkcries I don't know whether if you're being passive-aggressive or if you think it's cute lmao
1:31 yuyuko wearing every dress out there speedrun 100% glitchless
Omg your right
I need a madoka magica remake but all the witch layers look like this. The nightmare fuel.
Oh.... Oh no. That... is a geniunely fantastic idea and I need this in my life, but at the same time I dread it completely.
This is probably what you feel after seeing straight into Reisen eyes
Yo, the transition from 1:50 - 1:51 was prrtty good.
Background wallpaper material
Touhou artists love to portray characters in their own unique variations of their outfits, especially art for doijin circles. Every frame is a unique design for each character.
5:16 WHO IS THIS GUY XDDDDDDDDDDDDDDDDDDDDDDDDDD
Maybe young ZUN
Kinda scary how coherent Yukari looks at 3:03, feels like she tampered with reality.
1:24 I like how Flandre makes an extremely scrungy face instead of a grin.
Her jaw broke-
This literally feels like a dream, it's so drippy and so good
I'm still surprised as to how am I even able to remember my dreams clearly without any distortion whatsoever .w.
I think it's just your mind trying to make sense of the dream.
I know it's a spelling mistake but that doesn't make it less true, Touhou characters are dripped out fr fr. That Gensokyo Fashion is fantastic.
1:10 cup turns into liquid and splashes
0:20 this transition was so smooth
1:22 i love how she turns into a sillouhete (how do you spell it?) nodding on the the orignal, where the colors turn inverted .
2:19 i find it interesting how the characters appear to be on a stage in a concert due to the lighting. its a nice touch.
2:54 whatever the a.i. has going on with this character and their eye is very cool.
3:41 the way the water contrasts and is bright compared to everything else is a very nice touch. love it!
4:03 nice water effect and transition
4:43 i love the expressions and transitions.
Silhouette originates from French so its spelling is a little unusual
@@НепТор Sounds like bag-
The frame before she becomes a silhouette is both gold and nightmare fuel
0:28 it’s still heard that she hasn’t reached the house yet
We just not gon talk about how sick af 1:48 to 1:54 transition is? That was fire.
3:20 gotta speedrun my Essay in 2 hours be like:
1:21 this part is so underrated and it's literally the best in my opinion
i love th frame right before it turns into a silhouette how the face is all distorted
Yeah edgy cringe scarlet
Mi uxor se le plus belle..... 😭
Kind of scary to see how much AI can do.
0:45 this anime girl is growing up lmao
2:46 AYY REISEN WITH THE SNIPER RIFFLE
This is exactly how an acid trip looks like.
I once did, it really felt like this lol
i like how the AI sorta put the prismriver sisters onto a stage
this could be so so awesome with some more human intervention and better direction... some little additions here and there, getting things more clear and nice,
this even makes me as an artist feel distressed at being unable to actually intervene with the AI's works... not really changing it, but doing some overall little details.
in the end, the program intelligence is only reflecting the human mind and works behind them
it's fun to think there's not really a "real" boundary discerning the ones who created the AI (and the ones who ended up participating) and the final result of the program... yukari would be proud.
2:34 this feels like a distorted lucid dream
3:01 - 3:03 and 3:07 - 3:09 My favorite part of this ai creation
1:19 did the expression just break the AI?-
2:40 nah how did koishi get there
4:04 Damn what a nice transition though
Water physics broke, god please fix
0:22 bro literally walkibg in the sky
Tbh when i saw the thumbnail i thought i'd see some mega epic version of bad apple made by some good ai, but uh, i didnt expect it to look like... yknow.
imagine how good this could look if someone put a month into this, it'd probably be a masterpiece
4:51 was just amazing
is just me that feel the brain getting delayed while you is watching?
transition at 2:56 is actually really good
Yeah, prob the best 30 seconds
it is
02:50 excuse me, but what is that
AYo
Ikr😭😭😭
It's a small baby coming out of a girls thing
1:01
I really like that part, it's like she's saying: "No, No, No!"
this is how a 4D world would look like
0:57 easily my favorite scene it flows so well, AND it generates patchouli herself before the transition, dunno why but that seems perfect to me
I love how the characters don't really turn around, they just morph
Say what you will, but these transitions are still smoother than my pickup lines
4:37-4:50 okay that’s really good
This is one of the more mundane things I'm actually ecstatic to see people do with AI. It feels like a broken, senseless fever dream, but a masterpiece at the same time.
Or maybe I'm just high
Some of the frames in this video is wallpaper worthy
I HOPE THIS GET FAMOUS THIS IS REALLY COOL!
Hundreds of permutations of every character. Every changing accessory taken from some fan artist
The random guy at the end: 🕴️
2:01 for a split second you can kinda see Ellen from Touhou3 on the left
This feels like trying to change domething in a dream
I would never be the same again after this...
Wow. This looks so cool!
This feels like a fever dream, but the years watching Bad Apple running on multiple platforms has trained my brain to recognize the patterns. Absolutely amazing
0:15
Marisa is breaking speed limits on air
3:01 is the best transition for me
best thing i've watched in a while. dunno why, but i suddenly have my creativity and motivation back. haven't had that for i think 5 years now
Marisa, Sakuya, Yuyuko, and Mouku all look really good.
"Hey bro, what did you dream about?"
....
Well...
This is a perfect representation of waking up from a really badass dream and then trying to go back once you lay back down.
regular lsd trip
My eyes and head hurt!
Thank you.
this entire process is wild but there is just something, about low fps animations that make them better than high fps animations. it just looks cooler
Those upskirt shots though
i would pay so much money to see this rendered properly these transitions are insane
It feels like i'm seeing this from every reality at once
また新しい時代が始まった感じですね～！
AI特有のなめらかで溶けるような絵柄がぴったり合います。とてもドラマチック！
BRO MARISA WAS RIDING A NUKE THERE FOR A SECOND
0:53
No one gonna talk about the random apple logo?
Well, as a long time Touhou fan, this was jaw dropping to watch. Excited to see what more AI will bring us
You literally visualized how dreams go in our heads, gorgeous and immense simultaneously.
In a way, this is sort of a weirdly apt reflection of individuals' representations of each character. Everyone puts slightly different details into their art, so when it tries to render a specific character it is like we are viewing them through a kaleidoscope of individual poses and details from every alternate reality that exists. While it's sometimes kind of odd to see an individual pose turn around suddenly or take on a different silhouette the character remains identifiable. It is very interesting and beautiful.
This feels like when a character goes through a portal and we see multiple slightly different versions as they walk
Bro this is so trippy and it feels like I’m dreaming but why does the sensation feel so good??
I like how I keep seeing the Apple logo just flash for a few milliseconds lmao
1:49 is really fuckin cool
0:23 Mr. AI : "Sorry ma'am my database contains only panty shots"
It's like I'm watching a dream of some sort. You understand what is happening, Everything changes so quickly, there is no clear picture and you can not really remember a clear picture
Basically what dreams feel like when being half remembered at the point as you wake before being processed by your rational mind into something somewhat coherent.
This looks like some sort of glitchcore
This feels like a fever dream
02:55 Nezuko ?
3:55 did the audio just got bass boosted? 😂
It is supposed to be bass boosted
So this looks horrifically scattered right now...
But this is what we said about Dall E mini like 3 years ago.
AI made some of them damn thicc
0:45 this part was totally wrong but still was cool to see.
3:09 small Tenshi needing a chair made me crack up.
This is actually really cool
my last 2 brain cells during the exam thinking what to put in the essay:
Closest thing to seeing it fully animated in color :)
This feel like a fever dream 😮
Honestly, i would love to see this style expanded upon.