Faithful Where You Look: Anchor-Based Evaluation Hides Text Fabrication in Image-Conditioned Video Generation

arXiv:2611.00079 · cs.CV · Submitted 2026-10-18 · Read on arXiv

cs.CV

Submitted: 2026-10-18

Updated: 2026-10-18

Comments: 18 pages, 2 figures; includes supplementary material (S1-S11). Code and data: https://github.com/geoywang/refanchor-bench

Code: https://github.com/geoywang/refanchor-bench

License: http://creativecommons.org/licenses/by-sa/4.0/

The gist: Image-conditioned video models are used to show real places, so whether their output stays faithful to the place they were shown is a practical question.

Terminology

Abstract

Image-conditioned video models are used to show real places, so whether their output stays faithful to the place they were shown is a practical question. The text benchmarks we found that check fidelity against a reference do so at an anchor - an annotated region, typically the sign the task is about. We show that this choice hides most of the failure. We conditioned four models from three vendors on 48 street-level references of Tokyo storefronts and measured text fidelity at the annotated sign and in the periphery, the other text the reference covers. At the anchor, all four models are close to a no-model control (anchor text fidelity 0.90-0.94 against 0.996) and are not reliably separated from one another. In the periphery, 36-54% of generated text regions have no counterpart in the reference, against 1.6% for the control, although every prompt forbade new readable text. Two blinded raters agree on which regions are fabricated (kappa = 0.83-0.96), the result held in preregistered rounds on two vendors whose videos the measure had never seen (42% and 56%), and VBench-I2V's reference-based consistency scores show no evidence of tracking it in our sample. Most fabricated text is legible-looking text where the reference shows none or cannot be read - a form that string-matching metrics discard, which is why our own first measure missed it. A preregistered Latin-script replication in London missed its bar in a first round and met it in an independent second round. We release code, annotations, per-video scores, ratings and every preregistration. Code and data: https://github.com/geoywang/refanchor-bench

Sources

Related papers