You know we love bot-generated content. We’ve used various bots to create front-page stories, and to write a better Ace Attorney script than Shu Takumi, but we haven’t dabbled too much in the wild and wonderful world of Pokémon.
But Max Woolf has. Woolf is a data scientist at Buzzfeed, and has even created his own AI generation tools for creating bot-written text. One day, he woke up and decided to “feed” a bunch of Pokémon images to a bot — “a fine-tuned ruDALL-E” that’s not available to the public, because of its unique specifications and user-unfriendliness — to create brand new, almost-believable Pokémon.
And the internet went wild. Let’s first take a look at Woolf’s creations:
It’s weird, isn’t it? You can definitely tell which Pokémon were the main influences for some of them, and others — if you squint — look like the real deal. Others are less convincing, with Picasso faces and too many limbs, but then again… that’s pretty on-brand for Pokémon.
What’s that? You want more? Very well:
Unsurprisingly, these lovable abominations have spawned plenty of fan art. People are attempting to bring some order to the designs, and not all of them need that much work to make them believable.
The results of Woolf’s work are honestly pretty impressive — to teach a bot to make images look like the source material is hard, because bots don’t understand aesthetics. Most of the time, bots will try to smoosh everything together — like trying to combine the first 20 Google image results into one picture — but Woolf’s tweaking is the low-key star of the show, here. Clear features and defined faces are a large part of what make these Pokémon designs so close to the real thing, and that takes a LOT of specific instructions for bots.
But here are my six. I’m a Bot-Type Gym Leader. Fight me and my army of weirdos if you dare.
Any suggestions for names, types, and backstories for these bargain-bin reject Pokémon? Could you see any of them making it into the next game? Can you identify any of the parts? Tell us in the comments!