Video-based source camera identification (V-SCI) task in digital forensics relies on the extracted photo response non-uniformity (PRNU) patterns that deteriorate when the electronic image stabilization (EIS)—designed to reduce handshake motion of videos in modern cameras—is activated. We present a data-driven approach to exploit PRNU signals derived from EIS video via “device-specific” neural networks implemented with a novel PRNU image training and transfer learning strategy. Our technique is motivated by a discrepancy between how PRNU is used in real-world scenarios and existing methods’ design of SCI. Specifically, a forensics examiner in the field often has possession of the device in question, which is a source of richer device-specific features besides just the reference PRNU used in current SCI methods. We benchmarked the proposed method using EIS video sequences from the well-known VISION dataset and UDAYTON22VSCI, a new dataset we collected containing video sequences from 53 modern smartphone cameras, which confirms the advantages of our approach over state-of-the-art SCI methods. Speaker(s): Nicholas Hopkins, Virtual: https://events.vtools.ieee.org/m/339057