STORYANCHORS: Generating Consistent Multi-Scene Story Frames for Long-Form Narratives

TL;DR

StoryAnchors tackles the challenge of long-form, coherent multi-scene narrative video generation by decoupling high-level narrative planning from frame synthesis. It introduces a bidirectional story frame predictor, a multi-event labeling pipeline, and a progressive three-stage training regime built on a text-to-video backbone, enabling editable and expandable story frames with strong temporal coherence. Empirical results show StoryAnchors outperform open-source baselines on consistency, narrative quality, and diversity, with performance approaching GPT-4o in narrative richness and coherence. This framework provides a scalable, controllable foundation for future research in story-driven video generation and understanding, with practical potential for editable long-form content creation.

Abstract

This paper introduces StoryAnchors, a unified framework for generating high-quality, multi-scene story frames with strong temporal consistency. The framework employs a bidirectional story generator that integrates both past and future contexts to ensure temporal consistency, character continuity, and smooth scene transitions throughout the narrative. Specific conditions are introduced to distinguish story frame generation from standard video synthesis, facilitating greater scene diversity and enhancing narrative richness. To further improve generation quality, StoryAnchors integrates Multi-Event Story Frame Labeling and Progressive Story Frame Training, enabling the model to capture both overarching narrative flow and event-level dynamics. This approach supports the creation of editable and expandable story frames, allowing for manual modifications and the generation of longer, more complex sequences. Extensive experiments show that StoryAnchors outperforms existing open-source models in key areas such as consistency, narrative coherence, and scene diversity. Its performance in narrative consistency and story richness is also on par with GPT-4o. Ultimately, StoryAnchors pushes the boundaries of story-driven frame generation, offering a scalable, flexible, and highly editable foundation for future research.

Figures (8)

• Figure 1: The story frames generated by StoryAnchors exhibit high narrative quality and subject consistency across multiple scenes.
• Figure 2: The framework of StoryAnchors, with event prompts, condition frames and embeddings.
• Figure 3: Pipeline of Multi-Event Story Frame Labeling.
• Figure 4: Comparison of story frames generation. The frame sequences are generated by various models, including (a) StoryAnchors, (b) StoryDiffusion, (c) One Prompt One Story, (d) VideoGen-of-Thought, and (e) GPT-4o.
• Figure 5: Comparison of Open-Source Methods Across Five Evaluation Metrics (Evaluated by Gemini). This table summarizes the performance of StoryAnchors and other baseline methods under five evaluation metrics assessed by the Gemini system. The values in parentheses indicate the number of frames generated by each method.