K
K-2SO
Text
2501.00700
Generative models now synthesize photorealistic faces, and detectors that mine visual artifacts generalize poorly because of domain shift between their training categories and fine-grained facial tes…