This prototype visualizes all stories contributed by users during a small exhibition of the Macunaíma (formerly known as storyboarder) project in Prague.
All the stories started with the same prompt ("Create a story about Macunaíma, an AI parrot, that solves mysteries in the city of Prague"). Therefore, we are setting food in the fictional woods that grow from the interaction of this premise and GPT 3.5. Stable Diffusion is used for the images, adding a multimodal layer to the story.
The diverging paths are represented as a tree, each step in the story is a different node. The layout of the tree is computed with the dot algorithm of graphviz, for an easy representation that avoid overlaps and smooths out the visualization, removing edge crossings where possible.
The paths in the tree diverge and then join again together, using a system of clusters. In order to generate such clusters, we simply took all the strings of all the paths at a particular timestep and asked ChatGPT to group them together according to some semantic or thematic criteria, and name the clusters.
In the visualization, these clustered nodes are represented adjacent, and the title provided by ChatGPT represents the name of the cluster. While we are aware that this might not be a conventional method for clustering strings, it seemed fitting for the task at hand.
A few options for interactivity are left to the user:
Hovering over a node shows a zoom in on the hovered node, including image and text description.
Clicking on a node adds it to a panel on the right side. This is meant to allow users to craft their own storyboard by picking between nodes in the tree. They can also change the orders of the stories in the storyboard via drag and drop.
The slider allows users to change the amount of nodes displayed in the tree.