AIBS a web-based streaming service where AI uses original video cuts with algorithmically generated audio scripts to stitch together a bite-sized episode that fits the specifications of the user. The video below provides a tutorial.
Professors: Jenny Rodenhouse, Todd Masilko
"spock goes west"
I chose to feature a mash up of Star Trek and Westworld on the landing page because they represent a marriage between artificial intelligence and fan fiction, laying the foundation for this project. In the midst of my contextual research, I learned that the first recognized fan fiction was written for Star Trek. The strong diagonals in the header and logo are a reference to “slash”, the most iconic genre of fanfic.
I coined new terms for this project. This allowed me to highlight both how it is new and novel, yet still familiar and relatable for users. Tossing around words like "algorithm" would over-complicate a relatively simple user flow.
These images are from a text-to-image generator. They show the capabilities of such a model; there is a strength in the concrete, especially scenery. More complex topics become hazy for the model. I was immediately curious about how a machine would begin to interpret more abstract aspects of the human experience.
This project was initially inspired by running the TV show Gilmore Girls through a facial recognition model. I loved the idea of an AI attempting to understand relationships between characters given objective data. For instance things like the amount of time spent in visual proximity, types of objects in scene, statistical similarity between faces, etc.
1. Experiments to understand capabilities (runway) 2. Copy / pasting 5+ seasons of scripts to train a text-gen model 3. Use objects in scene to generate scripts (people, props) 4. Run scripts through Amazon Polly, manipulating ssml code 5. Cut together with video clips in After Effects
First, because having this service would be cool. And it's only possible with AI; in this instance, the AI acts as the ultimate fan. It knows every line and every character from every show its ever seen. With that knowledge, it can reference scenes at breakneck speed and create new content from old.
Content classification relies on tags. As spins are created, tags are automatically generated by the AI (show, character, genre), however before the spin gets posted, the user has an opportunity to edit the tags. This provides the AI with feedback so that over time its understanding of a fairly abstract concept like "genre" could grow and expand.
I especially love the idea of the AI cutting together a comedy and making "joke", then placing a laugh track. What kinds of jokes would an AI tell? How would its "sense of humor" morph in relation to the spin's popularity?
I let the content guide the interface. White space and a narrow color palette allow for bold imagery and mismatched type. I pulled proprietary letters from major networks for the logo and navigation titles, and paired them with one of the most universal and disavowed typefaces; arial. “Album” art for each episode mixes multiple images together in a slightly broken and repetitive fashion that mimics machine vision.
user flow insight
Initially I focused on the creation aspect of the flow, imagining users would spend lots of time within the spin creation tool. However I realized that like all popular content platforms, the reality is that most people will spend more time browsing than actually creating. Realizing that the flow would be more like the second model than the first, I went back and gave more time and attention to the browsing & remixing experience.
[recycling & covid-19]
Today, more than ever before, we are seeing a trend in recycling media and making old content new again. In the midst of this pandemic, while all production work is halted, a platform like AIBS answers a big need for something new. Within the creation tool, while a user is picking shows to spin, there is a small upload button, suggesting that users can actually become a part of their favorite shows and be featured in the spin.
i could go on forever.
I would love to talk about this project any time; please don't hesitate to reach out.