A Neural Network is Generating Really Convincing Fake POKÉMON

Look, we love Pokémon as much as the next nerdy website, but even we’ll admit that some of the creatures didn’t take a ton of human creativity to come up with. Birds like Pidgey and Spearow are just birds, while Klekfi is literally a floating set of keys. Sometimes, it feels like a computer could come up with new Pokémon without much trouble, and actually, it turns out that there’s a neural network that’s doing just that (via Kotaku).

Research scientist Janelle Shane decided to test a neural network and see if it could learn how to generate Pokémon names and abilities. It turns out that after not too long, the neural network became pretty darn good at coming up with convincing new creatures. Here are some examples of what it generated:

Quincelax
Abilities: Sturdy, Secene Grace
Hidden ability: Tunged LeusTortabool
Ability: Healy Stream

Strangy
Abilities: Wharmwbra, Darp
Hidden ability: Magic Guard

Staroptor
Ability: Stench
Hidden Ability: Stick Hat

Stangute
Ability: Banger
Hidden Ability: Drang

Tyrnakine
Ability: Beak Eye

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

Ronch
Abilities: None

Mawuh
Ability: Rum Power

Those sound pretty legit, right? What would cement them as being super realistic is if we could actually see what they look like. That’s where artist Lauren “Iguanamouth” Dawson comes in: She caught wind of the post and came to the same conclusion we did, so here are her renderings of the nine “Fakémon”:

Ronch is literally a bottle of Hidden Valley ranch dressing and Mawuh is just Agumon with a handgun, and we’re OK with that. Hit up the comments and let us know which of Iguanamouth’s designs is your favorite.

[brightcove video_id=”5175420990001″ brightcove_account_id=”3653334524001″ brightcove_player_id=”rJs2ZD8x”]

Images: iguanamouth