A fun little AI art widget named Text-to-Pokémon lets you plug in any name or description you like and (you guessed it) generate a Pokémon matching your prompt.

The model’s output isn’t flawless, but it’s incredibly entertaining all the same. You can try punching in the names of celebrities or politicians (see “Boris Johnson” and “Vladimir Putin” in the image above), or just general descriptions of the sort of Pokémon that would tickle your personal fancy (the one below is my “skeleton priest”).

(A little tip for using the tool more efficiently: make sure you select “4” from the “num_outputs” drop down menu to get four images per prompt rather than just one.)

The model is the work of machine learning researcher Justin Pinkney, who’s built a number of visual AI tools and resources. Notably, this particular model, Text-to-Pokémon, is adapted from a much bigger and much more powerful AI art generator named Stable Diffusion. While rival programs like DALL-E and Midjourney are locked down, Stable Diffusion is open source, making it easy for others to fiddle with its output. And that’s exactly what Pinkney did, fine-tuning the system using a database of Pokémon to create this little tool.

A quick search on Twitter shows people have been using Text-to-Pokémon to make all sorts of mash-ups, including Goku, Sonic the Hedgehog, and Jesus H. Christ himself. (Who, I’m going to say, is a dual Psychic/Ghost-type, known for his signature move “Resurrection.”)

In a thread on Twitter, Pinkney goes into a bit more detail on how he made the tool.

“Stable Diffusion is a great generalist model, but getting a certain style of output is pretty tricky, it usual needs some serious ‘prompt engineering’ (which I am rubbish at),” he says. “Fine tuning the model itself is an easy approach to focus on just what you want, if you have some data. I fine tuned the original stable diffusion on a Pokemon dataset.”

This is the big benefit of releasing open source AI models like Stable Diffusion: people come up with fun little tools like this. But it’s worth remembering that open source also has its downsides, and pretty much anyone is able to use Stable Diffusion to generate violent and sexual imagery, or misinformation and non-consensual pornography. You can read more about these tradeoffs and why Stable Diffusion creators’ released the model as it is right here.