# Monet Myrtle

  • What: A deep learning model that translates text into horrifying images. I didn't train this model, I just used it to generate awful pictures for my podcast.
  • When: 2021
  • Who: Me

# Model

Monet Myrtle is a VQGAN+CLIP (Vector Quantized Generative Adversarial Network and Contrastive Language-Image Pre-training) model that takes text as input and generates an image based on that description. I have worked with GANs before but I did not train this model. I'm actually not sure who should be credited for this piece of work but I used a Google Colab notebook created by Katherine Crowson to generate the images.

Most of the text inputs were suggested by listeners of my podcast Paging Mr Potter.

# Examples

I have only included a few examples below but there are more on our Instagram page as well as some time-lapse videos of the images being generated.

# "Vernon Dursley in the style of Picasso"

image


# "Gina Rinehart at an open mic night"

image


# "Chernobyl by Monet"

image


# "Alien fairytale"

image


# "Ron Weasley in The Matrix"

image


# "Tokyo Underwater"

image


# "Gringotts by Salvador Dali"

image


# "Watercolour of the Australian outback"

image