An exploration in synthesizing pornography
These images are all generated by a computer. No part of them are real photos. In more detail, they have been generated by a machine learning (ML) model which has been shown real pornography images and learns how to make its own. Even more specifically, it is a Generative Adversarial Network (GAN).
We're working on computer-generated pornography: images, gifs, eventually videos and even more eventually 3d/interactive content. Inspired by a friend's suggestion, we thought it'd be fun to share this early peak at this potential new approach.
The underlying technology, GANs, just isn't that good yet at complex large shapes like fingers, people and vaginas. Also, to save on resources, this model has been trained on 4 GPUs for 4 days (it's estimated that the most recent state of the art GAN took around $60,000 worth of GPU training to get its crisp high resolution outputs).
Awesome, more is coming!
We think so. There is one major benefit, that no people are harmed or trafficked in its production. There are other negative considerations, such as displacing the jobs of people currently in the adult film industry, and what affect [different kinds of] pornography have on the viewer. It's an on-going discussion and exploration.
Yes! We've got to the point where we need to break new ground in GAN technology, and to do that we need to hire more researchers and use a lot of computational resources. We also want to launch a consumer app, offering highly personalized pornography. Get in touch if you're interested in investing.