Directing the Narrative: A Finetuning Method for Controlling Coherence and Style in Story Generation

TL;DR

This paper introduces a novel two-stage framework for story visualization that improves coherence, character identity consistency, and style adherence in generated imagery, leveraging a new attention mechanism and preference optimization.

Contribution

It proposes Group-Shared Attention for intrinsic consistency and Direct Preference Optimization for aligning outputs with human standards, advancing story visualization quality.

Findings

Achieved +10.0 in Character Identity Score

Achieved +18.7 in Style Consistency Score

Outperformed baselines on ViStoryBench benchmark

Abstract

Story visualization requires generating sequential imagery that aligns semantically with evolving narratives while maintaining rigorous consistency in character identity and visual style. However, existing methodologies often struggle with subject inconsistency and identity drift, particularly when depicting complex interactions or extended narrative arcs. To address these challenges, we propose a cohesive two-stage framework designed for robust and consistent story generation. First, we introduce Group-Shared Attention (GSA), a mechanism that fosters intrinsic consistency by enabling lossless cross-sample information flow within attention layers. This allows the model to structurally encode identity correspondence across frames without relying on external encoders. Second, we leverage Direct Preference Optimization (DPO) to align generated outputs with human aesthetic and narrative standards. Unlike conventional methods that rely on conflicting auxiliary losses, our approach simultaneously enhances visual fidelity and identity preservation by learning from holistic preference data. Extensive evaluations on the ViStoryBench benchmark demonstrate that our method establishes a new state-of-the-art, significantly outperforming strong baselines with gains of +10.0 in Character Identity (CIDS) and +18.7 in Style Consistency (CSD), all while preserving high-fidelity generation.