Gaming Reviews, News, Tips and More.
We may earn a commission from links on this page

Computer Learns To Create Its Own Pokémon, And They're Great

We may earn a commission from links on this page.

You can train a Neural Network—basically a machine that learns—to do all kinds of things. The best thing, though, is to create a list of its own fake Pokémon. Please welcome Stangute, Tortabool and Ronch to the Pokédex.

“I decided to train [a] neural network to randomly generate Pokemon names and abilities based on this list as a training set - and found that it was good at generating Pokemon”, writes research scientist Janelle Shane on her blog. “Annoyingly good at it - by the time it had gone through the training set 50 times, it was already fluently plagiarizing Pokemon word for word.”

Here’s a list of some of the pocket monsters it has come up with:

Abilities: Sturdy, Secene Grace
Hidden ability: Tunged Leus

Ability: Healy Stream

Abilities: Wharmwbra, Darp
Hidden ability: Magic Guard

Ability: Stench
Hidden Ability: Stick Hat

Ability: Banger
Hidden Ability: Drang

Ability: Beak Eye

Abilities: Buttery armor, Shell Armor
Hidden ability: Weak armor

Abilities: None

Ability: Rum Power

We are all Ronch.

The only thing that would make the list better would be if we knew what these Pokémon looked like, so big thanks to Iguanamouth for doing just that and illustrating a bunch of them.


If you’re interested in the neural network and what else it can do besides Pokémon, Shane is also teaching it superhero names and a Lovecraft/cookbook mashup.