As part of our exploration into algorithmically generated film, we wanted to examine the possibilities of creating a modular system where a film could regenerate and change its own narrative based on both user and environmental signal. To that end, we created Good Adds Up, a celebration of all of the good that can happen when people come together on Facebook.

Using interactions between high-coefficient friends as a starting point, we trained the system to look for content from those relationships that had a high likelihood of positive sentiment based on data points like the ratio of reactions given, number of comments vs other similar posts, and people who are also tagged in that post. We then created a single, full length animated narrative that would house photos, videos, reactions, and comments from the data sets being surfaced.

To take things a step further, we created “holes” in the original animation and constructed alternate animations that could be slotted in their place - essentially, modular pieces of pre-constructed animation that could seamlessly house dynamically placed user content.

The result was a film that not only increased metrics across the platform, but drastically increased the sharing and virality of the films overall, as nearly every single person on the platform had a unique film they wanted to share.