Google DeepMind’s Fabula gives fiction writers hierarchical story control

Google DeepMind researchers, led by Piotr Mirowski, introduce Fabula – an interactive storytelling app designed as a 'sidekick' for fiction writers. Unlike generic AI writing tools, Fabula structures narratives hierarchically into scenes and beats, rooted in general narratological theory. Writers can generate, regenerate, and revise at both the script and story-plan levels, with the UI exposing the detailed plan alongside the final text. The system was built through participatory AI: design interviews and writing sessions with 42 casual and published writers, plus large-scale internal and external testing.

Fabula’s key innovation is its language model-based auto-evaluator, optimized on human experts’ preferences to improve story quality. The research interrogates four design choices: whether the auto-evaluator actually boosts quality, whether users want the exposed plan UI, how well narratology assumptions fit localised traditions (screenwriting, playwriting), and whether convergent iteration over the plan supports creativity. Critical feedback from writers highlighted tensions between AI guidance and artistic autonomy.

Beyond a writing tool, Fabula was used as a cultural probe in adversarial design to surface concerns about AI assistance in creative fields. The paper identifies new opportunities for writing feedback systems and interactive storytelling experiences. The team emphasizes that this is a research prototype, not a commercial product, but it points toward a future where AI co-writing tools balance narrative structure with authorial intent.