Stupid Idea For Trans Ppl: KMD

I just woke up from dead sleep to write this down because… you know what I want that I will actually work on this probs.

I wanna make a stupid simple deep fake tool so that you can see a picture of yourself as… well, yourself. How you see you. For me that’d be one thing, for someone else a whole other can o worms.

KMD, otherwise known as Kill My Dick, kek, would basically just be fed a shitload of stuff to eat and I’d have to figure out some of the tools… basically you’d feed it… oh I dunno, as many pictures as you wanna feed it I suppose, and you select in options about diff features and crap.

Not thinking like sim style app or nothin. Just like… I’m not doing surgely anytime soon… but mmmm… like a camera filter but better. I’d need it to auto match skin tones and shit, and hell I’m not even sure I have a good enough gpu to run ih.

But… idea!


What was that app that put an opposite gender filter on pictures? Maybe get hints from that how to work out your version if there’s code for it available.

I think I’m a little in over my head tbh but I’ll look into it

that depends which middle ware/how lazy

but basically seed it with pre hrt, 6 months, 1 year etc pictures?

I was thinkitg just feed as many images as possible.

the giga accel might help you if ya want the card it would be faster then any of ur gpus

1 Like

Maybe idk. I want it to work mostly.

1 Like

< dunno pretty sure most of the time is like pattern recognition or something

so would be like show it pre transition post transition pictures(as categories for the same person) and after billion pairs of that data, it would try to recreate that based on your input picture(prolly much easier to train only mtf, or ftm transition with their own data sets, then have it try to detect if its male or female and which way they were transitioning based on the input image)

What would be easier is running 2 separate algorithms for one program and have it model that seperately, with later tune ups being efforts to merge the two into one and then add more to it slowly.

That takes an assload oftimeand mole ppl thn I care to talk to in a day.

nah prolly be similar program, just a copy of same deal but trained with the other data set/a switchable data set, but ye same deal, need ass ton of images and in correct naming/organization to feed it before and after