Title: Characterizing the MrDeepFakes Sexual Deepfake Marketplace
Authors & Designations:
- Catherine Han, PhD – Stanford University
- Anne Li, PhD – Stanford University
- Deepak Kumar, PhD – University of California, San Diego
- Zakir Durumeric, PhD – Stanford University
Source: arXiv preprint arXiv:2410.11100 [cs.CY], originally submitted Oct 14 2024; revised Jan 30 2025 (arxiv.org, arxiv.org)
APA Citation:
Han, C., Li, A., Kumar, D., & Durumeric, Z. (2024). Characterizing the MrDeepFakes sexual deepfake marketplace. arXiv:2410.11100 [cs.CY].
🔍 Background & Objectives
Sexual deepfake content has surged online, driven by motives such as sexual gratification, harassment, and exertion of power in intimate relationships. MrDeepFakes has emerged as a leading open-access marketplace for creating and distributing such content. This study systematically examines its economic structure, targeted individuals, and community behaviors to understand contemporary deepfake production practices (arxiv.org).
🧠 Methodology & Data
- Quantitative: Scraped public metadata covering 43,000 videos featuring 3,800 unique individuals and over 1.5 billion views as of November 2023 (arxiv.org).
- Qualitative: Analyzed 830 paid request forum threads to uncover buyer motivations and pricing. Reviewed 25+ tutorial threads to assess technical needs for deepfake creation (arxiv.org).
📈 Key Findings
- Targets: 95.3% of videos depict female actors and musicians; over 14% target non-public figures despite rulebook “celebrity only” policy. About 2.4% involve rape, abuse, or humiliation scenes (arxiv.org).
- Economics: Average price for custom deepfakes is $87.50, ranging from $1 to $1,500. Pricing is influenced by target dataset quality, video complexity, and turnaround time (arxiv.org).
- Technical Barriers: Required assets include 3 K–15 K facial images per target. Community employs open-source toolkits, Google Colab, and cloud GPUs. Forums provide peer support and reference academic research (arxiv.org).
🧭 Implications & Discussion
- Weak Rule Enforcement: Platform guidelines regarding celebrity-only targets and prohibition of exploitative content are frequently ignored.
- Novel Harassment Incentives: Buyers cite nonconsensual gratification, power dominance, and a perceived sense of “contributing art” as driving motivations—expanding known typologies of abuse (arxiv.org).
- Dual-Use Concern: The community openly shares sophisticated technical methods, referencing machine learning literature—suggesting a crossover between academic knowledge and abusive use .
📌 Conclusion
Han et al. firmly establish MrDeepFakes as the largest public sexual deepfake marketplace. Their work reveals its rapid growth, economic dynamics, technical sophistication, and the urgent need for targeted policy, technical mitigation strategies, and further research to combat nonconsensual intimate imagery online. This study serves as a foundation for academics, platform designers, and policymakers addressing AI-driven sexual abuse.
Original Abstract (verbatim):
“The prevalence of sexual deepfake material has exploded over the past several years… Our work uncovers little enforcement of posted rules…, previously undocumented attacker motivations, and unexplored attacker tactics for acquiring resources to create sexual deepfakes.” (arxiv.org)
Follow JSTOR.ONLINE