Orange County, CaliforniaGitHubLinkedInEmail
James Skorupski
← Research
AI & Interactive Narrative

StoryCanvas & Beyond StoryCanvas

Visual Comic-Storyboard Authoring & Runtime Analysis for Plan-Based Interactive Drama

OrganizationUC Santa Cruz — Expressive Intelligence Studio (EIS)
RoleLead Researcher & Architect (Ph.D. Dissertation Research)
Period2009–2012
StoryCanvas visual plot fragment editor interface
StoryCanvas prototype showing comic-panel plot fragment authoring with visual preconditions and story actions.

StoryCanvas: Novice-Friendly Authoring of Plan-Based Storyboards

StoryCanvas is a comic- and storyboard-based interactive story authoring and generation environment built on the author-goal-centric UNIVERSE planning model introduced in Wide Ruled.

Creating generative, branching stories is fundamentally akin to programming a reactive state machine—a daunting barrier for writers and narrative designers without formal computer science training. Inspired by the expressive visual shorthand of comic books and film storyboards, StoryCanvas replaces text-heavy rule tables with direct visual manipulation of characters, environments, precondition badges, and comic panels.

To help authors manage large, combinatorial narrative spaces, StoryCanvas pairs visual authoring with continuous real-time feedback on the topology, coverage, and branching factor of the underlying plan-based story world.

Beyond StoryCanvas: Runtime Story-Space Analysis & Mixed-Initiative Planning

As part of my Ph.D. advancement research at UC Santa Cruz, I designed a suite of next-generation architectural extensions (Beyond StoryCanvas) addressing scalability and mixed-initiative AI collaboration in large interactive narratives:

1. Passive Story-Space Exploration & Coverage Heatmaps

Any rich generative story contains a deep hierarchy of Author Goals, Plot Fragments, and interdependent state preconditions. To give authors an immediate mental model of how their rules interact at runtime, a background analyzer continuously simulates playthroughs as the author edits—rendering a reachability heatmap directly over the goal hierarchy to expose dead ends, starving branches, or over-dominant plot fragments (effectively continuous runtime code-coverage analysis for narrative planners).

2. Structural Hierarchy Compression & Procedural Panel Modifiers

  • Subtree Collapse & Similarity Compression: Automatic and manual folding of large Author Goal subtrees based on structural and semantic similarity across Plot Fragments.
  • Visual Storyboard Pane Modifiers: Precondition-driven actions that dynamically alter panel framing (camera pan/zoom composition) and color grading to reflect shifting character emotions or world state without requiring duplicate art assets.

3. Mixed-Initiative “Delinearization” via Social Physics (Comme il Faut)

To bridge traditional linear storyboarding with branching AI planning, I designed a mixed-initiative workflow powered by the Comme il Faut (CiF) social-physics engine (used in UCSC’s Prom Week by Josh McCoy):

  1. Step A (Linear Draft): The author sketches a straightforward linear sequence of story panels.
  2. Step B (Planner Query & Exploration): The author highlights a subsequence of panels and asks the planner to “delinearize” it. Using a corpus of social rules and dramatic arc heuristics (rising tension, conflict, denouement), the planner searches the space of character traits and relationships to suggest candidate branching Plot Fragments.
  3. Steps C & D (Selection & Branch Integration): The author tunes character traits or accepts one or more AI-suggested variations, which are automatically woven into the Author Goal hierarchy as playable branches.