Just one artificial intelligence could retire Arceus and Ken Sugimori himself at the same time. In the last year and a half, AI has undergone an unprecedented evolution thanks to OpenAI. This company has already shown that artificial intelligence applied to text and images has only just begun. Using the capabilities of Dall-E, OpenAI’s imaging artificial intelligence, a computer engineer has been able to create a software that spawns fake pokemon.
Now you won’t be able to get them all
Generate Fakemon using AI is a project of Liam Eloie, a computer engineer expert in machine learning. Liam is a huge fan of Pokémon, and believes that at the rate at which The Pokémon Company generates monsters, there shouldn’t be many years until they run out of ideas. His project was born as much out of his passion for the franchise as it was for artificial intelligence.
Although his website is only four months old, the engineer assures that he has been working on this project for years. It all started when generative adversarial networks (GANs) became fashionable. Liam tried generate Pokémon images using these models so advanced at the time, but rarely got decent results. As we said at the beginning, AI has drastically evolved since 2020, and this has allowed the engineer to fulfill his objective.
The real turning point in its AI capable of spawning fake Pokemon was Dall-E, just over a year ago. The big difference between this OpenAI product and the others is that it allows combine generative models with our natural language. From this moment on, Liam Eloie could input an input to his network ‘Generate a grass-type Pokémon’ and his artificial intelligence could understand the concept and generate an image according to the request.
How does Generate Fakemon using AI work?
Liam Eloie’s website is very simple. He has two tabs: one that we can use free Y another more advanced with functions premium. In the first we can select the type of a Pokémon and generate as many images as we want. If any interest us, we can save it to consult it later. We can also choose that the entire process is random.
On the other hand, the website has an advanced mode that allows you to generate double type pokemonas well as legendary. You can only generate them as 3D models, but in an upcoming update, the ‘Ken Sugimori’ style will be selectable. This second mode is paid, and Liam uses it to finance development.
In previous versions, this app worked giving as input directly a pokemon. However, the engineer decided to modify this option so as not to seek legal problems with The Pokémon Company.
As usual in the world of AI, you need to hit generate several times to get an interesting result. Despite this, the program achieves very good resultsand Liam often shares his users’ best creations on his app’s official Twitter from time to time.
A project that can be significantly improved with Dall-E 2
The creators of fangames of Pokémon surely they have just seen the open sky with this app. But… What if we said that this application could take an even bigger jump in quality next year? As we already told you a few weeks ago, OpenAI announced Dall-E 2.0, and the differences with respect to the original version are gigantic. So far only a lucky few have access to the API of this new version, but we are sure that as this limitation is removed, we will see projects like Generate Fakemon using AI with even more amazing quality.
If you are curious about how Dall-E 2 works, here is a video in which the youtuber Dot CSV tests it live. We recommend that you skip the video little by little so that you can appreciate the power that this new AI from OpenAI has.
