Apologies to Women of Color
As you can see, I've started posting galleries again. XHamster added some AI specific options to video posting, so it looks like they are getting a handle on AI content. Not quite enough for me to get monetized videos posted, so I can start earning my own AI computer, sadly.
Galleries kind of lags videos in AI features. They don't have specific check-marks to specify it is AI like videos, but the staff seem to be better trained now, and I don't get the continual "which performer is this" queries. Unfortunately they don't have good fidelity when they reject a gallery. All I get is that it is rejected, not why, or which picture. I end up making my best guess, removing a few pictures, and then trying again.
Sadly, most often this involves removing pictures of women of color. The reasons are a bit complicated.
First, we all know AI is biased in its representation. But one of my goals here is to correct for that bias, and by intentional use of prompting present smut representative of all racial and body types. I mean, I love skinny Asian chicks as much as the next person, but I think they are already very well represented and choose to concentrate on other types.
Second, we know that AI impersonation is a problem, and there are plenty of bad actors out there making images of actual people without permission. This is really hard to police and so XHamster mainly enforces this by requiring AI smut to be "cartoony".
The problems come where these two requirements interact. If I just ask my AI to draw some pictures of cartoony African American women, I tend to get pretty boring (and occasionally offensive) stereotypes. Often caricatures with big-hips and massive afros. To counteract this I have a vocabulary of terms I use to elicit better and more representative images of African American women. (Especially for their hair!) And this works pretty well, until I try to also make them "cartoony". Unfortunately, the AI generators I've been using aren't very well trained on a comprehensive set of graphic illustrations of African American women. And when I start adding the terms to make them look more accurate, what I get is them also looking more real. And this then leads to XHamster rejecting galleries.
Oddly enough, AI doesn't seem to have the same problem with Indian women.
So, you may notice my galleries may be a bit lighter in content of dark skinned women. And now you know why. I will continue to struggle on, seeking out new AI generators with better training data. Some day I'll be able to have my own rig and will train it the way I want it trained. But until me an XHamster work out how to do revenue earning videos, that's going to be a while.
Funny addendum:
I was please to be proactively contact by XHamster support with respect to some issues I was having. After a few exchanges though, I worked out that this contact was being facilitated by AI. Now, I think AI can be super helpful in providing better support, and I use it for that in other, unrelated projects. It just got a bit surreal talking about AI generated content with someone using AI generated content to reply to me. :-)
Galleries kind of lags videos in AI features. They don't have specific check-marks to specify it is AI like videos, but the staff seem to be better trained now, and I don't get the continual "which performer is this" queries. Unfortunately they don't have good fidelity when they reject a gallery. All I get is that it is rejected, not why, or which picture. I end up making my best guess, removing a few pictures, and then trying again.
Sadly, most often this involves removing pictures of women of color. The reasons are a bit complicated.
First, we all know AI is biased in its representation. But one of my goals here is to correct for that bias, and by intentional use of prompting present smut representative of all racial and body types. I mean, I love skinny Asian chicks as much as the next person, but I think they are already very well represented and choose to concentrate on other types.
Second, we know that AI impersonation is a problem, and there are plenty of bad actors out there making images of actual people without permission. This is really hard to police and so XHamster mainly enforces this by requiring AI smut to be "cartoony".
The problems come where these two requirements interact. If I just ask my AI to draw some pictures of cartoony African American women, I tend to get pretty boring (and occasionally offensive) stereotypes. Often caricatures with big-hips and massive afros. To counteract this I have a vocabulary of terms I use to elicit better and more representative images of African American women. (Especially for their hair!) And this works pretty well, until I try to also make them "cartoony". Unfortunately, the AI generators I've been using aren't very well trained on a comprehensive set of graphic illustrations of African American women. And when I start adding the terms to make them look more accurate, what I get is them also looking more real. And this then leads to XHamster rejecting galleries.
Oddly enough, AI doesn't seem to have the same problem with Indian women.
So, you may notice my galleries may be a bit lighter in content of dark skinned women. And now you know why. I will continue to struggle on, seeking out new AI generators with better training data. Some day I'll be able to have my own rig and will train it the way I want it trained. But until me an XHamster work out how to do revenue earning videos, that's going to be a while.
Funny addendum:
I was please to be proactively contact by XHamster support with respect to some issues I was having. After a few exchanges though, I worked out that this contact was being facilitated by AI. Now, I think AI can be super helpful in providing better support, and I use it for that in other, unrelated projects. It just got a bit surreal talking about AI generated content with someone using AI generated content to reply to me. :-)
1年前
