StoryBench: A Multifaceted Benchmark for Continuous Story Visualization

TL;DR

StoryBench is a comprehensive benchmark designed to evaluate and advance text-to-video generation models across multiple tasks, emphasizing realism, consistency, and adherence to prompts, with a focus on multi-task evaluation and human assessment.

Contribution

It introduces a new multi-task benchmark with annotated datasets for evaluating text-to-video models and provides guidelines for human evaluation and insights into automatic metrics.

Findings

Training on story-like data improves model performance.

Current models struggle with complex story generation tasks.

Guidelines help standardize human evaluation of video stories.

Abstract

Generating video stories from text prompts is a complex task. In addition to having high visual quality, videos need to realistically adhere to a sequence of text prompts whilst being consistent throughout the frames. Creating a benchmark for video generation requires data annotated over time, which contrasts with the single caption used often in video datasets. To fill this gap, we collect comprehensive human annotations on three existing datasets, and introduce StoryBench: a new, challenging multi-task benchmark to reliably evaluate forthcoming text-to-video models. Our benchmark includes three video generation tasks of increasing difficulty: action execution, where the next action must be generated starting from a conditioning video; story continuation, where a sequence of actions must be executed starting from a conditioning video; and story generation, where a video must be generated from only text prompts. We evaluate small yet strong text-to-video baselines, and show the benefits of training on story-like data algorithmically generated from existing video captions. Finally, we establish guidelines for human evaluation of video stories, and reaffirm the need of better automatic metrics for video generation. StoryBench aims at encouraging future research efforts in this exciting new area.