ReDiStory: Region-Disentangled Diffusion for Consistent Visual Story Generation

TL;DR

ReDiStory is a training-free method that enhances visual story generation by reorganizing prompt embeddings to better preserve subject identity across multiple images, reducing semantic interference.

Contribution

It introduces a novel inference-time prompt reorganization technique that explicitly decomposes and decorrelates embeddings to improve identity consistency without additional training.

Findings

Improves identity consistency metrics on ConsiStory+ benchmark.

Maintains prompt fidelity while reducing cross-frame interference.

Operates without modifying diffusion models or requiring extra supervision.

Abstract

Generating coherent visual stories requires maintaining subject identity across multiple images while preserving frame-specific semantics. Recent training-free methods concatenate identity and frame prompts into a unified representation, but this often introduces inter-frame semantic interference that weakens identity preservation in complex stories. We propose ReDiStory, a training-free framework that improves multi-frame story generation via inference-time prompt embedding reorganization. ReDiStory explicitly decomposes text embeddings into identity-related and frame-specific components, then decorrelates frame embeddings by suppressing shared directions across frames. This reduces cross-frame interference without modifying diffusion parameters or requiring additional supervision. Under identical diffusion backbones and inference settings, ReDiStory improves identity consistency while maintaining prompt fidelity. Experiments on the ConsiStory+ benchmark show consistent gains over 1Prompt1Story on multiple identity consistency metrics. Code is available at: https://github.com/YuZhenyuLindy/ReDiStory