This AI Can Draw Human Penises Full-Time, Thanks To Google

Dick-drawing AI

What does it take for an AI to learn how to draw a penis? For a project called 'DICK-RNN', it initially required 10,000 pictures of human male genitalia.

This project is actually a fork of SKETCH-RNN, but customized for penises. As described by the researchers of the former in a paper, the goal is to "train machines to draw and generalize abstract concepts in a manner similar to humans."

The AI was trained on stroke-based vector drawings, capable of handling unconditional generation of vector images.

And Google's role here in helping the development of this 'dick-sketching' DICK-RNN AI is when it open-sourced the Quickdraw data set.

It was this data set that the DICK-RNN developer used to train the AI draw penises.

From Studio Moniker's Quickdraw-appendix project:

"In 2018 Google open-sourced the Quickdraw data set. 'The world's largest doodling data set'. The set consists of 345 categories and over 50 million drawings. For obvious reasons the data set was missing a few specific categories that people seem to enjoy drawing. This made us at Moniker think about the moral reality big tech companies are imposing on our global community and that most people willingly accept this. Therefore we decided to publish an appendix to the Google Quickdraw data set."

Then in June of 2019, the developers of Quickdraw released the Do Not Draw a Penis project to "collect inappropriate doodles from people who are not willing to stay within the moral guidelines set by our social network providers."

This initiative helped them collect another 250,000 doodles.

And here, DICK-RNN was trained with 10,000 images of human penises from the Quickdraw-appendix data set, processed via incremental RDP epsilons to fit most penises within 200 steps.

This was later increased to 25,000 images with a maximum length of 300 steps.

Dick training data set
Training samples from the data set

In other words of saying it, DICK-RNN is simply SKETCH-RNN but trained with penises.

The AI is able to generate possible ways to finish drawing penises. The AI can also encode existing sketches into a latent vector, and generate similar looking sketches, conditioned on the latent space.

The goal for this application is to encode a crude, poorly sketched drawing, and generate more aesthetically looking reproductions that resemble penises.

"Note that I used the raw version, not their simplified version, since the dicks were more detailed. The processed dataset that is compatable with sketch-rnn's strokes (no pun) is in this repo," said the DICK-RNN developer.

"I also believe that 'Doodling a penis is a light-hearted symbol for a rebellious act' and also 'think our moral compasses should not be in the hands of big tech,'" said the developer on the project's GitHub page.