XPlainVerse: A Million-Scale Benchmark for Explainable Deepfake Detection
2026-07-03 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors created XPlainVerse, a large collection of one million real and fake images designed to improve deepfake detection by not only spotting fakes but also explaining why an image is fake. They developed a method called Edit-Check to ensure the fake images truly show the intended edits, which helps train better explanations. Their dataset offers two types of explanations: detailed ones for experts and simpler ones for everyday users. They also introduced new ways to measure if explanations really point to the manipulated parts correctly. Overall, their work aims to make deepfake detection more trustworthy and understandable.
deepfake detectionexplanation generationvisual artifactsdataset benchmarkimage manipulationmulti-stage filteringreasoning fidelityEntityScoreEvidenceScorehuman-centered explanations
Authors
Abhijeet Narang, Kartik Kuckreja, Shreya Ghosh, Muhammad Haris Khan, Jianfei Cai, Abhinav Dhall
Abstract
As deepfake detection models increasingly produce natural language explanations, their reasoning often remains weakly grounded in visual artifacts, limiting reliability and user trust. Existing benchmarks mainly evaluate classification accuracy, overlooking whether explanations reflect the actual manipulations. This gap hinders progress toward deployable, explainable deepfake detection systems. To this end, we introduce XPlainVerse, a large-scale benchmark designed for joint deepfake detection and human-centered explanation. XPlainVerse comprises one million real and manipulated images, pairing authentic images from five established sources with forgeries generated by twelve off-the-shelf image editing and synthesis models. We further propose a multi-stage filtering pipeline, Edit-Check, to verify if manipulations satisfy their intended edits, enabling reliable reasoning supervision at scale. Beyond dataset scale, XPlainVerse provides two complementary explanation styles: technical explanations for expert analysis and simplified explanations optimized for non-technical users. To evaluate explanation quality beyond surface similarity, we propose novel metrics, EntityScore and EvidenceScore, that measure reasoning fidelity by checking whether explanations correctly identify manipulated entities and visual evidence. Human annotations on 2,000 explanation pairs validate our dataset quality against human judgment. We believe XPlainVerse will establish grounded explanation quality as a measurable dimension of deepfake detection and support scalable research on trustworthy, interpretable models.