Super Epic Happy Fun Time

Backstory

While working at Pixar, people would ask me what I was working on all the time (mostly because I was in so many different departments and working on so many random projects). I would joke that I was working on a “Make Movie” button and that soon my entire job would be automated.

With the rise of AI and the ability to synthesize images, speech, and text I decided maybe it was finally time to build the “Make Movie” button.

Predating the rise of text-to-video models such as Sora, this project aimed to utilize the state-of-the-art AI tools freely available at the time, and add a dose of human touch to them.

Setup

The design aesthetic of the world resembles the prolific art of Keith Haring because his characters lack teeth, fingers, and the other details AI particularly struggles with.

The show is staged as a multi-cam sitcom which allowed me to set the camera positions ahead of time in each scene.

The show is in Japanese because the text-to-speech models at the time sounded uncanny when speaking English. I figured a lot of people are used to watching Japanese anime and I could just subtitle it.

Execution

Each episode starts as a .json file which defines the name of the episode, the location, the background audio track, the characters in the episode, each of their lines, and the camera position for each of their lines.

Then a script would distribute all of the tasks on Deadline and the entire episode would render.