Deepfake detection in the era of large models

With the continuous advancement of artificial intelligence, Deepfakes have evolved from single-modality synthesis into complex generative forms involving visual, auditory, and textual media. The emergence of large multimodal models (LMMs) has significantly enhanced the capability to generate forged content, while simultaneously bringing unprecedented opportunities and challenges to the task of forgery detection. This paper presents a comprehensive review of recent progress and technological evolution in forgery detection under the background of large models. It surveys relevant research outcomes over the past three years and summarizes recent multimodal forgery detection datasets. On this basis, we conduct an in-depth analysis of the potential and challenges of LMMs in terms of detection performance, hallucination, judgment accuracy, and fairness. We analyze the underlying causes and propose future solutions to address these issues. Finally, the paper explores future trends in forgery detection technologies, including the increasing complexity of forgery information, the value of traditional techniques, explainability, and technological adversarial dynamics.