[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/ic/ - Artwork/Critique

View post   
File: 388 KB, 1447x836, kek.png [View same] [iqdb] [saucenao] [google]
6445490 No.6445490 [Reply] [Original]

what do you guys think of this method to fight back against AI? surely us humans would be faster at shitting out crap like the pic to flood the database they are scraping from than it takes a pajeet to type his hundreds of proompts and picking the best one?

/beg/s, now's your chance to post on artstation and be recognized for your hard work!

>> No.6445499

Kek, Id love to see that happen just for shit and giggles

>> No.6445500

>>6445490
AI doesn't "unlearn" things and I believe current models use arts and photos from some time ago. Models are also composed by billions of trained data, 100K bad arts or even more wouldn't hurt it.

>> No.6445502
File: 462 KB, 800x682, 164568546546.png [View same] [iqdb] [saucenao] [google]
6445502

>the reason ai art is so good
gave me a chuckle

Keep going though
there is potential for zozzles

>> No.6445528

>>6445490
nah we good. Algorithmic disgorgement will do just fine

>> No.6445532

so this is what desperation looks like.
lmao.
lol.

>> No.6445542

>>6445490
Simply run all scraped images through something like that
https://github.com/christophschuhmann/improved-aesthetic-predictor
and then filter out all images below certain score

>> No.6445551
File: 45 KB, 512x512, witheredHD.jpg [View same] [iqdb] [saucenao] [google]
6445551

>>6445490
It wouldn't work, AI doesn't train in real time.
It is possible to "poison" the data base but you have to actually have control of it and also willingly train on bad data.

>> No.6445690

>>6445490
This depends on how each trainer handles their database.
For stable diffusion they use the LAION database, but they don't use the complete one, I'm pretty sure they use LAION-Aesthetics, which basically is a machine learning program going through the original database and picking the ones it deems the more "aesthetic", so the end result is that it ends up with high quality paintings like renaissance ones and artistic photography, but its a machine learning program so it would also leave in modern and contemporary art if it finds the textures and technique of those "aesthetic"

https://laion.ai/blog/laion-aesthetics/

>> No.6446231

The only thing this does is proving that the anti-AI pearl clutchers are as clueless to what they are dealing with as the pajeets

>> No.6446671

>>6445490
I think the damage it does to ai is so minor as to be unnoticeable, and in the process you make your portfolio look like shit
So doing so honestly damages yourself more than ai

>> No.6447424

>>6445490
>pajeet
Are they really the ones flooding AI art everywhere, or is this just a meme?

>> No.6449154
File: 179 KB, 972x1727, 1671857435933810.jpg [View same] [iqdb] [saucenao] [google]
6449154

>>6447424
>is it just a meme