ML Security PapersTowards Interactive Deepfake Analysis
Lixiong Qin 1,2, Ning Jiang 1, Yang Zhang 1, Yuhan Qiu 1, Dingheng Zeng 1, Jiani Hu 2, Weihong Deng 1
Published on arXiv
2501.01164
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
First interactive deepfake analysis system using MLLMs that supports natural language querying for detection, forgery technique classification, and fine-grained artifact description of face images.
DFA-GPT
Novel technique introduced
Existing deepfake analysis methods are primarily based on discriminative models, which significantly limit their application scenarios. This paper aims to explore interactive deepfake analysis by performing instruction tuning on multi-modal large language models (MLLMs). This will face challenges such as the lack of datasets and benchmarks, and low training efficiency. To address these issues, we introduce (1) a GPT-assisted data construction process resulting in an instruction-following dataset called DFA-Instruct, (2) a benchmark named DFA-Bench, designed to comprehensively evaluate the capabilities of MLLMs in deepfake detection, deepfake classification, and artifact description, and (3) construct an interactive deepfake analysis system called DFA-GPT, as a strong baseline for the community, with the Low-Rank Adaptation (LoRA) module. The dataset and code will be made available at https://github.com/lxq1000/DFA-Instruct to facilitate further research.
Key Contributions
- DFA-Instruct: a 127.3K image / 891.6K QA-pair instruction-following dataset for deepfake analysis constructed with GPT assistance
- DFA-Bench: a benchmark evaluating MLLMs on deepfake detection, classification, and artifact description tasks
- DFA-GPT: an interactive deepfake analysis system built by LoRA fine-tuning an MLLM, serving as a community baseline
🛡️ Threat Analysis
Core contribution is detecting AI-generated deepfake images/videos — determining whether faces are real or synthetically generated and describing forgery artifacts. This is AI-generated content detection, a primary ML09 use case.