Do Generative Priors Align with Human Naturalness Perception?
Abstract
Visual generative models are trained to capture the probability distributions of natural images, yet whether their native priors reflect the regularities governing human perception of image naturalness remains an open question. Here, we probe these priors through native prediction errors across 25 open image and video generators. Because raw single-image losses are dominated by scene content and visual complexity, we evaluate directional loss differences using content-preserving, paired relational interventions that selectively disrupt facial configurations or physical illumination consistency while limiting changes in low-level image statistics. Across both domains, these loss differences reproduce human-like selective sensitivities and tolerances, capturing the classic Thatcher effect on faces and shape-dependent responses to illumination inconsistencies. Notably, these loss differences reliably track continuous gradations of human naturalness judgments across individual stimulus pairs (peaking at r = .84 on faces and $.64$ on physical scenes) and retain unique human-aligned signals even after controlling for feature distances from frozen vision encoders and standard image quality metrics. We also find that while overall sensitivity to these violations broadly covaries with human alignment across models, the two systematically decouple along denoising schedules, with alignment peaking earlier than sensitivity, revealing that human-like naturalness judgments dissociate from generic violation detection. Together, these findings demonstrate that learning visual distributions yields generative loss landscapes that capture distinct aspects of human naturalness perception.
Related
- Top-down semantic predictions align phonetic neuronal dynamics in human superior temporal gyrus
- Projected Energy Matching for Generative 3D Priors
- Mapping the neural basis of the perception of “flow of time”: interval timing and mental time travel in the human brain
- Gait kinematics dynamically tune optic flow motion-in-depth perception during human locomotion
- Intrinsic fluctuations modulate thalamic gating during the emergence of human conscious perception
BibTeX
@misc{fukiage2026generative,
title = {{Do Generative Priors Align with Human Naturalness Perception?}},
author = {Taiki Fukiage},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.09928},
eprint = {2610.09928},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.09928},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Do Generative Priors Align with Human Naturalness Perception?
AU - Taiki Fukiage
PY - 2026
DA - 2026/10/07
PB - arXiv
DO - 10.48550/arxiv.2610.09928
UR - https://arxiv.org/abs/2610.09928
AB - Visual generative models are trained to capture the probability distributions of natural images, yet whether their native priors reflect the regularities governing human perception of image naturalness remains an open question. Here, we probe these priors through native prediction errors across 25 open image and video generators. Because raw single-image losses are dominated by scene content and visual complexity, we evaluate directional loss differences using content-preserving, paired relational interventions that selectively disrupt facial configurations or physical illumination consistency while limiting changes in low-level image statistics. Across both domains, these loss differences reproduce human-like selective sensitivities and tolerances, capturing the classic Thatcher effect on faces and shape-dependent responses to illumination inconsistencies. Notably, these loss differences reliably track continuous gradations of human naturalness judgments across individual stimulus pairs (peaking at r = .84 on faces and $.64$ on physical scenes) and retain unique human-aligned signals even after controlling for feature distances from frozen vision encoders and standard image quality metrics. We also find that while overall sensitivity to these violations broadly covaries with human alignment across models, the two systematically decouple along denoising schedules, with alignment peaking earlier than sensitivity, revealing that human-like naturalness judgments dissociate from generic violation detection. Together, these findings demonstrate that learning visual distributions yields generative loss landscapes that capture distinct aspects of human naturalness perception.
ER -