[Discussion] A Comprehensive Benchmark for Story Visualization: ViStoryBench

[ViStoryBench]

Hi Story2Board Team,

Congratulations on your excellent work on Story2Board! We were very impressed by your training-free approach to generating truly expressive storyboards. Your focus on compositional diversity and narrative pacing is a much-needed perspective in this space, and the proposed Latent Panel Anchoring (LPA) and Reciprocal Attention Value Mixing (RAVM) mechanisms are very clever. 💡

We also noticed your introduction of the Rich Storyboard Benchmark, which is a fantastic contribution, especially with its focus on narrative expressivity and the new Scene Diversity metric.

In a similar spirit of advancing standardized evaluation in this field, our team has developed ViStoryBench, a comprehensive benchmark suite designed to evaluate story visualization models across a wide range of criteria. We believe it could be a valuable and complementary resource for your work.

• 🎬 Diverse and High-Quality Dataset: It includes 80 curated multi-shot stories from diverse sources (literature, film, folklore) with 344 characters and 509 manually curated reference images.

  📊 Comprehensive & Automated Metrics: We offer a automated metrics to thoroughly evaluate models on character consistency (CIDS), style similarity (CSD), prompt adherence, and aesthetic quality.

  🏆 Public Leaderboard: We maintain a leaderboard with results from over 18 SOTA methods and commercial systems, providing a broad context for comparison.

We believe that evaluating Story2Board on ViStoryBench could provide an insightful, complementary perspective, allowing for a direct comparison against a wide array of existing methods on a different set of diverse narratives.

Since your method is training-free, it's an ideal candidate for integration. We would be very happy to try adapting it to ViStoryBench ourselves and will gladly update the results on our public leaderboard.

We have open-sourced the entire benchmark, evaluation code, and a public leaderboard to track progress.

• Project Page: https://vistorybench.github.io/
• Paper: https://arxiv.org/abs/2505.24862v3
• Code & Dataset: https://github.com/vistorybench/vistorybench

We look forward to seeing more amazing work from your team and hope we can work together to advance the field of story visualization!

Best regards,

The ViStoryBench Team