Unrelated but I once scraped r18.com for Japanese Adult Video (JAV) videos and created an artificial intelligence that could generate JAV titles, here's a couple of the best ones:
- Undercover Footage Of Handsome Students Masturbate While Their Pussies Get So Sensitive They Can't Get Caught... and Their Families Fuck To Their Fetishes vol. 40
- Half Dosage Aphrodisiacs x Wakaba Onoue Aphrodisiac Revenge Porn - Incestual Sex With A Beautiful Barely Legal Girl An Adorable Fate Ai Uehara Kurea Hasumi , Hikaru Konno
- Encyclopedia Of 2 Mind Blowing Orgasms! Endless Pissing! Pooping + The Ultimate Ass & Golden Shower - Best Of Collection Aika Yumeno
- I'm A Cherry Boy (Not!) I Went To Visit My Grandma, But This Is The Story Of How My Wife(Age 34) Got Fucked By Her Co-Worker(Age 20) At Me, So She Came To My Room And Asked, "Show Me Something Nasty," And We Started To Skip Practices But Then We Fucked" vol. 1
I’m A Cherry Boy (Not!) I Went To Visit My Grandma, But This Is The Story Of How My Wife(Age 34) Got Fucked By Her Co-Worker(Age 20) At Me, So She Came To My Room And Asked, “Show Me Something Nasty,” And We Started To Skip Practices But Then We Fucked" jumped the shark after vol 4 imo
For some reason "Half Dosage Aphrodisiacs" is the funniest thing to me.
Give me the strongest aphrodisiacs you have. No that's too strong.
Not sure yet, gotta see what colab can handle. Each panel is 32x32px, but that was basically just so I could finish before going to sleep last night. I'm going to do some optimization today and see if I can get the resolution up, but honestly I kind of like the low-res look.
the original panels are 175x200px, and downscaled to 32x32 the model took probably 2 hours to train. I wrote the project from scratch so its not impossible that if we just feed it into stylegan or something we might get something approaching readable text. But will garfield be as dummy thicc as he is in that last panel?
https://imgur.com/a/fCDd9LO there is a good Gon in the third panel of this one
Do you have any of this up on github? One of my friends really loves Garfield (the cat, not so much the comic lol), and I'd love to send her some ML-generated garfields.
Or even if you could just point me in the right direction regarding how to replicate the project myself, like have you found any good introductions to doing this I could reference?
Don't want to link my github because it's under my real name. I can send you the code if you like or point you in the right direction. How much coding experience do you have?
Understandable. If you could just point me in the right direction, I guess? I have a decent amount of coding experience, but I'm a bit rusty. I do some stuff with python and bash for work, used to work with C, but haven't in years.
Also - I have a single RTX 2070 Super, is that powerful enough to even run stuff like this? I worry I might not have the right hardware.
https://machinelearningmastery.com/what-are-generative-adversarial-networks-gans/ https://machinelearningmastery.com/how-to-develop-a-generative-adversarial-network-for-a-cifar-10-small-object-photographs-from-scratch/
I would start with these two links. If you understand python syntax then you're most of the way there. The first link basically just describes the concepts behind adversarial networks, and the second actually has code that should run fine on a 2070 if you want to use that as a springboard. If you run into hardware limitations, Google Colab is basically a free python notebook that will do all the gpu processing on their machines, with some limitations that can be removed for $10 a month (how long your sessions can last etc). The code on that page should run fine on the free tier though. Otherwise, just youtube GANs and watch talks and stuff. The computerphile video is a good place to start for someone who already knows a bit of math and programming. https://www.youtube.com/watch?v=Sw9r8CL98N0
Its eerie how much squinting at this hoping it will makes sense feels like looking at a comic your dad is showing you at breakfast while trying to pretend you're sober
why is OpenAI wasting their time on language research when they could dedicate their computing powers towards unlimited Garfield
If you can make a program that is able to recognize a certain kind of image (like a cat, dog or Garfield strip), you can pit it against another program that feeds the first program either a real image of that thing, or an attempt at generating one. The first program has to guess whether or not the image is real, and sort of tells the second program how it knew. Over time, the second program learns change things about its generated images that tip off the first program that the image is fake, and if you do that enough, you are left with a program that is really good at making images indistinguishable from the real thing. https://www.youtube.com/watch?v=Sw9r8CL98N0 Here is a good video that basically explains what I just wrote but better.
