1. StableDiffusion fine-tuned on Pokemon images
2. a basic RNN for Pokémon name generation
3. a basic OpenCV background-removal model.
The results can be occassionally excellent. Some recent good prompts from users are:
- automobile with wings
- water pokemon with two heads and amphibian legs
- jeff bezos
- phil collins
Other good previous prompts are provided as auto-complete options. My favorite prompt at the moment is 'Willy Wonka Cat' because the model nails the combination of Gene Wilder's Willy Wonka outfit and a typical feline Pokémon form.
There's really no interesting technical innovation in this demo. It's just a hopefully interesting combination of stuff that already exists. It's become so easy to stick together ML models, often without training most or all of them yourself.
video demo: https://youtu.be/mQsMuM8d4Qc
cloud platform: https://modal.com
code: https://github.com/modal-labs/modal-examples/tree/main/06_gp...