EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation

Ruozhen He, Meng Wei, Ziyan Yang, Vicente Ordonez · 2026

Read EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation by Ruozhen He, Meng Wei, Ziyan Yang, Vicente Ordonez (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-15199.txt.

Introduction

Ruozhen He and three others offer EntityBench, a benchmark for entity-consistent long-range multi-shot video generation. Old evaluations use independent prompt sets with thin entity coverage and simple match metrics. EntityBench holds 140 stories and 2,491 shots from real narrative media, tracking characters, places, and objects across tiers up to 50 shots, with return gaps up to 48 shots. A three-pillar suite splits shot quality, prompt fit, and cross-shot match, gated by a faith check on entity looks.

Catalog record

CREATORRuozhen He, Meng Wei, Ziyan Yang, Vicente Ordonez
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE136593
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2605.15199

Browse the full library · About the Tower of Babel