PVLM: Parsing-Aware Vision-Language Model With Dynamic Contrastive Learning for Zero-Shot Deepfake Attribution

This research paper introduces PVLM, a new method for identifying which AI system created a deepfake (fake video or image of a person's face) by analyzing how well different generators preserve facial features. The approach combines vision-language models (AI systems that understand both images and text) with face parsing (analyzing individual facial components) and dynamic contrastive learning (a training technique that groups similar items together while separating different ones) to better recognize deepfakes from unseen advanced generators like diffusion models (AI systems that create images by gradually removing noise).