After experiencing the nine generations of " Pokémon ", players sometimes inevitably complain that developers just randomly draw words from their hats and observe what kind of chemical reactions they will trigger in the player group. For fans who have the mentality of "it's okay to do this", here is a unique artificial intelligence generator that can be recommended to everyone.
From Justin Pinkney, a machine learning researcher at Lambda Labs, just shared a "text to Pokémon" generator created by fine-tuning Stable Diffusion.
We have followed up and reported on the relevant progress of OpenAI's DALL-E 2 drawing AI many times, and "text to Pokémon" is also the same as it.
To achieve this goal, Pinkney carried out special description training on Pokémon for AI models.
For example, " Wonderful Frog Seed " (Bulbasaur) is described as "red-eye and green-skinned", while " Green Caterpillar " (Caterpie) is labeled as "red-nosed yellow-green toy".
Pinkney further explained on Twitter that once the model is fine-tuned, everyone can easily generate images of similar styles.
Friends who want to use this OpenAI text to image generator, just link your GitHub account and then you can easily get started.
In this process, you can switch the number of image outputs, the steps to denoising, and the size of classifier-free boot.
Interestingly, many netizens tried to enter their names into the "text to Pokémon" generation system, and the result was so funny.
