Speakers

Postcards from the Generative Edge: Investing in Visual Narrative Generation with The Marvel Method

To go on a journey is to experience a story first-hand, one that we can later tell others. Conversely, to invent a new story we must first imagine a journey for our characters to go on. With the narrator as a guide, the audience becomes a vicarious traveller along the same path. Since stories move characters through space and time in more-or-less goal-directed ways, it is natural to think of stories as journeys along a path, whether physical or emotional. We see this in the language we use to talk about narratives: we speak of stories filled with twists and turns, or of stories that lose their way or that get derailed or stuck in the weeds, or of page-turners that race at breakneck speed toward a thrilling conclusion. Time scales and geography can vary greatly across stories with the same underlying path structure. Homer’s Odysseus spent 10 years on his return from the Trojan war; James Joyce’s counterpart, Leopold Bloom, spent just a single day on his mock-heroic journey across Dublin, but his Ulysses mirrors all of the key story events of the original Odyssey.

A story-generating machine can send characters on a journey by first setting them on intersecting or overlapping paths toward their respective goals. It defines these paths with a starting point, an end point, and a shape, and gives characters the means and the motives to travel along them. Different media will treat the shape of a story more or less metaphorically. In comics, however, a character’s path is rendered quite literally, as a visual sequence of postcard-like panels that punctuate different stages of their journey. Comics vary widely in their sophistication and narrative complexity, but no young reader ever needs to be taught how to consume one. Although intuitively simple, the visual grammar of comics is also subtle enough to modulate the shifting pace and focus of a narrative, by allowing for variation in the size, shape and contents of each panel.

Comic, like films, both show and tell their stories, so when it comes to generating them, these responsibilities can be shared by creators with complementary skillsets. The “Marvel method”, a collaborative approach to comics creation pioneered in the early days of Marvel Comics by Stan Lee, Jack Kirby and Steve Ditko, flipped the script on how visual narratives are typically created. Conventionally, comic stories are written first and illustrated later, but using the Marvel method, stories are illustrated before they are written. After an initial plot discussion, the artist first maps out the narrative visually, to produce a path-like sequence of panels that the writer can later interpret and augment textually. When using the Marvel method to tell a tale, the postcards are sent before they are written, and they arrive all at once. The accompanying text is only written once the journey has already concluded.

The Marvel method was once a productive model for comics collaboration, but it is rarely used today. Too much of the creative burden rested on the artist’s shoulders, while too much of the credit went to the writer. Nonetheless, it remains a useful division of labour for automated comics generation, for here too we have machine collaborators with very different skillsets that must work together to show and tell their story. In this talk I will explore how Large Language Models can work side by side with visual story grammars, to tell stories that marry words with images and old with new AI paradigms.