No articles I can find yet but online chatter shows people talking about how tonight Grok is now heavily censored. Won't take uploaded images and put the woman in a bikini, etc. Won't make videos of women undressing into bikinis/underwear etc.
I tested it personally by uploading a pic of a woman and asking it draw her in a bikini and it was moderated. Uploaded to Imagine a woman already in a bikini and it refuses to make a video of literally anything (not even "woman reads a book", how sexist!)- even with no prompt it refuses. Asked it to draw a woman in a bikini and 1 attempt was moderated but another was generated.
All because Elon stuck his dick in crazy.
Yeah, but I feel like when they blew this issue up they barbra streisanded it though.
There are so many resources/tutorials to start this on your own. And while most, stress the potential legal issues of this, many people simply use randomly created people and create "personas". But even that kinda bothers me because chances are prolly higher than you think that whoever you randomly generate is going to have a likeness to someone in the real world.
I also figured, you know its prolly alot easier to spot the generated content if you know how it works.
The only thing that is stopping people doing this offline is if they have a computer video card older than 6 years or very low vram.
Using something like comfyui and then a text to image workflow like qwenedit to generate the initial image and then using a image to video workflow for wan2.2 to create some pretty decent stuff from the get go. Of course with the comfyui listed models it will do some basic topless stuff (for artistic vision ofc), it wont do much more than that, but plenty of user trained datasets available trained on stuff like onlyfans, not hard to swap the diffusers out. With wan though it seems like you basically have to create your own character loras or the details will shift alot in the scene, using something like ostris ai toolkit.
Im retarded and I figured a good chunk of it all out in a few days, if I can do it...I will offer a piece of advice, to do this involves installing hundreds of different packages, you need so many prereqs, python, cuda, git, are the major ones, do it on a fresh install of windows with no personal information on there, and just plan on nuking again when your done.
Im glad I dont have a bunch of personal photos on the cloud somewhere, all it takes is like 15 pictures, from what I read, to get a convincing model of someone. The thing is training loras takes massive vram, or massive time... so thats a little more out of reach for the average person unless your ok with putting your pictures into the cloud for a paid processing service(if you wonder where all the video card chips and ram are going). But all ive messed around with is wan2.2, ive read sdl1.5 can cut that time in half.
I really want to try and do some characters powered by neural nets in a video game, eventually, and i figured this is the first step towards that. Part of me is half tempted to try and make a short 5 minute show, robot chicken style or something.
Ill prolly die long before i accomplish anything though. :shrugs: