Dexter Studios Partners with KETI to Develop Deepfake Detection Solutions: “Acting as an ‘AI Red Team’ Identifying Vulnerabilities via Film-Grade Deepfakes”
Dexter Studios is taking the lead in developing core technologies to address deepfake crimes, which are rapidly spreading into daily life and emerging as a serious societal issue.
Comprehensive content studio Dexter Studios (206560, hereinafter “Dexter”) announced its participation in the government-backed project “Development of Active Detection and In-Depth Analysis Solutions for Complex Generative Deepfakes” led by the Korea Electronics Technology Institute (KETI) and sponsored by the Ministry of Science and ICT (MSIT) and the Institute for Information & Communication Technology Planning & Evaluation (IITP). The consortium also includes Korea Advanced Institute of Science and Technology (KAIST), the National Forensic Service (NFS), the Korean National Police University (KNPU), Safe AI, and Secure Machines Lab.
With the rapid advancement of generative AI technology, sophisticated complex deepfake contents combining face swapping, voice modulation, and background editing have surged. Single-modality models specialized in only audio, image, or video faced technical limitations with reduced detection accuracy when identifying high-quality new types of deepfakes.
The consortium aims to develop an active “Detection Agent” where AI independently analyzes the characteristics of deepfake content to select the optimal model, along with a “Multi-Detection Engine” integrating audio, image, and video detection capabilities. The core mission is to create an in-depth integrated solution that synthesizes multiple detection results to determine forgery while providing explaining evidence for its judgment. The developed solution will undergo real-world validation through the National Forensic Service and the Korean National Police University for performance enhancement.
For this project, Dexter will deploy its hyper-realistic fine-grained modification technologies and high-resolution post-processing control capabilities to generate and build dataset of commercial film-grade face-swapping and subtle alteration deepfakes. Utilizing this ultra-high-quality dataset, Dexter plans to test system limits, uncover vulnerabilities, and continuously enhance detection system performance against evolving deepfake generation methods.
Dexter has previously demonstrated exquisite execution in large-scale film, series, and OTT VFX pipelines—replacing faces and expressions of real actors, crafting precision volumetric-scan digital humans, and implementing AI-based re-aging (aging and de-aging). Dexter’s industry-leading image generation technologies are expected to directly contribute to developing high-performance solutions validated through law enforcement and forensic research institutions.
Jae-won Song, Head of Dexter’s R&D Center, stated, “Dexter has developed cinema-level image generation technologies capable of adjusting an actor’s lip movements to match multi-language dubbing or swapping faces while retaining the actor’s original emotional performance. Recognizing the risks posed by emerging generative AI deepfakes, we joined this project to act as an ‘AI Red Team’ that continuously exposes detection model vulnerabilities by synthesizing top-tier deepfake data.”
ⓒ Dexter Studios
