On May 11th 2020 PUFF was born

In the spring of 2020 (before chatGPT and text-to-image models existed) I trained a neural network on around 100 of my hand-drawn monster artworks.

The resulting GAN model could produce endless synthetic monster drawings in an echo of my style.

One of these stood out to me over the others: a pink monster with rabbit ears. I named her PUFF.

Over the next few years, PUFF became a vehicle for exploring the weird edges of AI and creativity — ideas around authorship, style, originality, data ownership, and what it means to ‘create’ something.

PUFF virtual being latent space walk

The evolution of PUFF:

With the help of the online community, PUFF evolved.

I built a new dataset using beautiful fan-made drawings of her, and trained a new version of the model. An AI trained not on my work, but on people's interpretations of a synthetic character.

This new model could produce infinite versions of PUFF.

PUFF data set images

Synthetic PUFF images

PUFF experiments

Beyond the flat image:

Using newer AI tools, I pushed PUFF into different dimensions — reimagining her across materials, scenes, styles, and more.

The project became a kind of sandbox for testing what’s possible and what is to come in this new AI future.

I started this project as a playful experiment into a new technology, but as AI has become more widespread and controversial, the project has become more charged. I don’t have all the answers, but PUFF has helped me think through and embody some of the issues.