DETECTING SUBTLE DEEPFAKE MANIPULATIONS

Summary from the TLDR AI newsletter

Leveraging CLIP's visual encoder and parameter-efficient tuning, this method robustly detects partial deepfake forgeries across diverse datasets with minimal model changes.