Researchers have developed MTOR, a novel system for detecting AI-generated videos by leveraging both visual and textual semantic information. The system identifies a phenomenon called temporal over-regularity (TOR) in AI-generated content, where videos exhibit excessive temporal consistency and reduced variability. MTOR integrates global visual data with text derived from video captions and specifically models TOR at multiple temporal levels. Evaluations across five benchmarks and 46 generator variants show MTOR achieving state-of-the-art performance against 16 existing methods, demonstrating robustness against various real-world perturbations. AI
IMPACT This research introduces a more robust method for detecting AI-generated videos, addressing the increasing challenge posed by advanced synthetic media generation.
RANK_REASON Academic paper detailing a new model for AI-generated video detection. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →