See what kinds of /d/ material which is obviously porn you can slip through the filter at IsItPorn.com
Post best results
I bet it works by modeling color patterns, as it seems extra bad at slime girls.
This gets fewer points for the higher score, but still surprised that it wasn't given 90+
>>7323617
Or it might have just a simple "weird color = not porn" rule and then much more rules for determining if pics containing skin tone are porn or not.
Finding false positives is probably more interesting.
>>7323622
Suicide and/or murder.
>>7323838
"Share amusingly unexpected IsItPorn.com results for /d/ material" probably would have been a better premise. If the thread catches enough interest to bear repeating I'll phrase the OP better next time.
I was actually surprised how often it was able to catch certain stuff that wasn't explicit but was porn. Didn't actually save any examples though because they wouldn't have matched thread. Oh well.
0.1%
>>7323614
Beta test our product for free the thread
A trap getting spitroast, safe.
Dickgirl getting boned. Safe.
Futa Harpy, safe.
Psychotic harpy goddess/demon of wind? Safe.
1/3
>>7326752
2/3
3/3
I find it hilarious the most objectionable one is the one with missionary sex and kissing.
http://randomguy664.tumblr.com/post/83918589700/
This is my new favorite fetish.
Trust the website, it knows dirty, dirty porn. Clearly I have not been fapping to the best stuff.
after trying a couple images, i agree that the ones showing more skin or skin colors are more likely to be considered porn, and with huge numbers too ( 95% + )
>>7326769
that's actually a pic made for people into the invisible people fetish.
very accurate
kek
The algorithm is looking for human skin tones more than anything else.
>>7329347
note the massive difference in percentage
this is right
>>7326734
Is that Funny Valentine?
let's not train ais to censor things we love
>>7325289
Well, it is censored.