How Poke Blend works is pretty easy to understand. So Poke Blend is an AI that has learned what 535 different Poke Man look like on a pixel by pixel level, more or less, & with that knowledge, we can ask it to show us what Poke Man A & Poke Man B would look like if combined. This is possible because the AI has learned what each poked man looked like based on its number. Its as if I were to tell you "please draw me #11" & you knew exactly how to make it with almost pixel-by-pixel accuracy. This is what Poke Blend is doing. So because it understands each of the 535 Poke Man as a mathematical representation, we can ask it to show us what a (11 + 357) / 2 Poke Man would look like & whala! & Yes, ideally if I weren't lazy, you'd be able to blend 3, 4, or as many Poke Man that you'd like to see what you'd get. 😜

Poke Blend was first trained through much trial & error using Neural Networks & tensorflow / keras machine learning libraries, & once it was trained to a point where it was good enough, it was then converted into a tensorflow json model to be used with JavaScript & HTML, which is how this site is possible. Also shouts to @DavidBates for his JavaScript help on this website.

Now of course these results aren't jaw dropping amazing, there is room for improvement, & I have some ideas, but I mostly wanted to make this site just to have an excuse to learn how to get my python models online, & to just see how users interact with this project. I am very interested.

If you would like more info on the ins & outs / behind the scenes of this projects, please check out my 4-Part Devlog below! Or feel free to shoot me a tweet over on twitter I am @jabrils_!