Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Can ChatGPT Detect DeepFakes? A Study of Using Multimodal Large Language Models for Media Forensics

Shan Jia, Reilin Lyu, Kangran Zhao, Yize Chen, Zhiyuan Yan, Yan Ju, Chuanbo Hu, Xin Li, Baoyuan Wu, Siwei Lyu

TL;DR

Problem: can multimodal LLMs detect DeepFakes without programming? Approach: evaluate GPT4V Vision and Gemini 1.0 Pro Vision on the DF3 dataset using targeted prompts that yield Yes/No, probability, and artifact-based explanations, with results averaged over multiple queries to compute $AUC$ and rejection rates. Findings: GPT4V achieves approximately 0.78–0.80 $AUC$ on raw and post-processed AI-generated faces, outperforming Gemini in this task, while real-face detection remains less reliable and largely driven by semantic cues rather than signal-level artifacts. Significance: demonstrates the potential of vision-enabled LLMs as intuitive forensics tools, while highlighting limitations (semantic reliance, post-processing robustness, API restrictions) and the need to combine prompting improvements with traditional signal-based detection for robust DeepFake detection.

Abstract

DeepFakes, which refer to AI-generated media content, have become an increasing concern due to their use as a means for disinformation. Detecting DeepFakes is currently solved with programmed machine learning algorithms. In this work, we investigate the capabilities of multimodal large language models (LLMs) in DeepFake detection. We conducted qualitative and quantitative experiments to demonstrate multimodal LLMs and show that they can expose AI-generated images through careful experimental design and prompt engineering. This is interesting, considering that LLMs are not inherently tailored for media forensic tasks, and the process does not require programming. We discuss the limitations of multimodal LLMs for these tasks and suggest possible improvements.

Can ChatGPT Detect DeepFakes? A Study of Using Multimodal Large Language Models for Media Forensics

TL;DR

Problem: can multimodal LLMs detect DeepFakes without programming? Approach: evaluate GPT4V Vision and Gemini 1.0 Pro Vision on the DF3 dataset using targeted prompts that yield Yes/No, probability, and artifact-based explanations, with results averaged over multiple queries to compute and rejection rates. Findings: GPT4V achieves approximately 0.78–0.80 on raw and post-processed AI-generated faces, outperforming Gemini in this task, while real-face detection remains less reliable and largely driven by semantic cues rather than signal-level artifacts. Significance: demonstrates the potential of vision-enabled LLMs as intuitive forensics tools, while highlighting limitations (semantic reliance, post-processing robustness, API restrictions) and the need to combine prompting improvements with traditional signal-based detection for robust DeepFake detection.

Abstract

DeepFakes, which refer to AI-generated media content, have become an increasing concern due to their use as a means for disinformation. Detecting DeepFakes is currently solved with programmed machine learning algorithms. In this work, we investigate the capabilities of multimodal large language models (LLMs) in DeepFake detection. We conducted qualitative and quantitative experiments to demonstrate multimodal LLMs and show that they can expose AI-generated images through careful experimental design and prompt engineering. This is interesting, considering that LLMs are not inherently tailored for media forensic tasks, and the process does not require programming. We discuss the limitations of multimodal LLMs for these tasks and suggest possible improvements.
Paper Structure (10 sections, 10 figures, 4 tables)

This paper contains 10 sections, 10 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Large Language Models
  4. DeepFake Faces: Generation and Detection
  5. Methodology
  6. Experiment Results
  7. Qualitative and Quantitative Results
  8. Ablation Studies
  9. Improvements
  10. Conclusion

Figures (10)

  • Figure 1: The overall process of using multimodal LLMs to detect AI-generated face images.
  • Figure 2: Which of these images are real and AI-generated? Answer:
  • Figure 3: Examples of evaluation data. 'SG2' stands for the StyleGAN2 model, 'LD' represents the Latent Diffusion model, and 'PP'ed' means post-processed data.
  • Figure 4: Examples of GPT4V for DeepFake face detection. Left: Results for AI-generated images from the $DF^3$ dataset ju2023glff. Right: Results for real faces from the FFHQ dataset karras2019style. The responses for AI-generated faces are labeled in pink, while those for the real faces are labeled in green. Both success (w/ ) and failure (w/ ) cases are shown.
  • Figure 5: Examples of Gemini 1.0 Pro for DeepFake face detection. Left: Results for AI-generated images from the $DF^3$ dataset ju2023glff. Right: Results for real faces from the FFHQ dataset karras2019style. The responses for AI-generated faces are labeled in pink, while those for real faces are labeled in green. Both success (w/ ) and failure (w/ ) are shown. We can see that even though some yes/no results are accurate, the supporting evidence is not.
  • ...and 5 more figures