mots - Latent Space

In 2022, or "a few years before AI-pocalypse" as some will call it later into the future, we were invited to participate at Metamersion: Latent Spaces, an event funded by Champalimaud Foundation.

The goal was to produce an art-science installation together with a team led by Zach Mainen consisting of a VR experience and a short film, focused on the future of AI in neuroscience.

For the short film, we figured, why not ask AI?

Using a few prompts and the promise of world domination, we convinced the machine to come up with some scene outlines. These, we combined into a working plot, which we fed back to the AI to give us a screenplay.

A few extra touches to prevent a financial disaster and our 8-page screenplay is now ready to be produced with the help of our non-AI but still very intelligent biological human friends from Zauberberg.

Fast forward to editing: we realized that the plot would be more impactful if the character sees an older version of herself in the mirror.

click to play

But, instead of waiting 50 years for the actress (Siori) to age gracefully, we trained our own AI model which is also capable of time travel. A few hours later:

click to play

The AI model is really versatile and can be used to generate a wide range of Sioris. Here are a few other aged Sioris: [1] [2] [3] [4] [5]

Or Cyberpunk Siori, or a Manga Siori

Or how about a crowd of Weird Sioris?

That's enough!! Sorry, Siori!

The vision with the cliff is supposed to be representative of something from the character's past. AI kept pushing us to think about climate anxiety, polar bears, old houses, childhood traumas, and chairs. We let the audience think about what these representations mean, but here are a few shots that AI created for us on top of the footage we captured in Portugal.

Almost no additional human work was needed on these shots, by the way.

Therefore, we thought it's a good idea to not release one, but many variations of the film - each variation adding meaning, and revealing a bit more about the character. Here are 20 variations of weird hands that the character is seeing:

click to play click to play click to play click to play

Having multiple variations of the short film allowed us to play them in sync on multiple screens in the same time, so that the audience could choose their own experience, almost like the protagonist, each AI therapy session at a time.

We're not sure if we achieved this, and rendering out and quality controlling 20 versions of the short film was the opposite experience of working with AI... But, again, in case this wasn't clear, the only purpose of this project for us was to prove that we know how to work with AI, so let us know where to send those job applications, despite what AI is saying!