A delightful little AI art widget anointed Text-to-Pokémon allows you to plug in any name or description you like and generate a Pokémon matching your prompt. However, they can only do action in your mind.
The model’s output isn’t perfect, but it’s incredibly entertaining. For example, you can try perforating in the names of celebrities or politicians.
The model is the work of machine understanding researcher Justin Pinkney, who’s built several visual AI tools and resources. Notably, this model, Text-to-Pokémon, is adapted from a much bigger and more powerful AI art generator named Stable Diffusion. While rival programs like DALL-E and Mid journey are locked down, Stable Diffusion is open source, driving it easy for others to fiddle with its output. And that’s precisely what Pinkney did, fine-tuning the system using a database of Pokémon to complete this little tool.
A brief search on Twitter reveals people have been employing Text-to-Pokémon to make all kinds of mash-ups, including Sonic the Hedgehog, Goku, and Jesus H. Christ himself. Pinkney goes a bit more about how he made the tool in a thread on Twitter.
“Stable Diffusion is a fantastic generalist model, but getting a certain output style is pretty tough; it usually needs some serious ‘prompt engineering,'” he states. “Fine-tuning the model is easy to focus on just what you want if you have some data. For example, I fine-tuned the original stable Diffusion on a Pokemon dataset.”
The significant benefit of releasing open-source AI models like Stable Diffusion is that people develop fun, petite tools like this. But it’s worth recognizing that open source also has its downsides, and anyone can use Stable Diffusion to generate violent and sexual imagery, misinformation, and non-consensual pornography.
Stable Diffusion is a text-to-image prototype assembled by teamwork between engineers and researchers from Stability AI, CompVis, and LAION.
It is based on a model named Latent Diffusion (High-Resolution Image Synthesis with Latent Diffusion Models). Unfortunately, the theoretical components are beyond the content of this article.
Because the generated photos are a close contest to Open AIs Dall-E 2, perhaps even satisfactorily, they open-sourced it with the code and the model weights.
Using Stable Diffusion AI
- Log in or Sign up to Hugging Face: Hugging Face – The AI community creating the future. They make the model you are about to operate, and they have given access to their Google Colab notebook.
- Now that you are logged in, you must get the access token you will employ later by going to the accounts’ settings at the top right edge.
- In settings, you will detect the option called Access Tokens. Initiate a new token.
- Offer your token a suitable name and choose the read role.
- Copy the induced token to the clipboard; you will need it later in the Jupyter notebook.
- Log in to the Google Colab. Once there, access the Stable Diffusion notebook.
- Enter the access token you replicated in the earlier step. Pause for the Login Successful message.
You are ready to employ the model! The notebook is relatively self-explanatory so that you can try various prompts directly.
You can never anchor at a perfect prompt without much trial and error. For instance, the AI will not render the same images with the exact prompt each time. Additionally, you ought to know a lot about the art world, the artists, and the additional styles used to be able to offer good prompts. True art understands how to inform the computer to draw!