Unsurface applies peer-reviewed forensic methods to detect, localize, and explain synthetic and manipulated visual media — for newsrooms, defense, intelligence, and law enforcement teams whose work depends on knowing what's real. Our core technology, Forensic Self-Descriptions, was developed at Drexel's Multimedia and Information Security Lab and published at CVPR 2025.
Identify synthetic images and video produced by diffusion models, GANs, and other generative systems — including generators Unsurface has never encountered during training. Zero-shot detection built on forensic traces, not on recognizing specific models.
Beyond real-or-synthetic: cluster imagery by the family of generator that likely produced it, even within a single case, and even when the source is unknown. Open-set attribution that scales with the generative ecosystem as new models appear.
Every analysis produces interpretable signals your team can inspect, discuss, and document. Not a black-box verdict — a forensic report designed for expert review, defensible in editorial debate and investigative work.
ROADMAP
Unsurface is actively expanding what forensic analysis can do — adding capabilities our founding partners have identified as the highest-leverage additions to their verification workflows.
Identify exactly where an image or video frame has been altered, not just whether it has. Turning "this looks off" into "this is what's off, and here."
In-house tooling that works alongside your team to trace the provenance of visual media and cross-reference public sources, accelerating open-source investigations that currently take days.
Face-swap and identity manipulation detection · detection of general video editing and manipulation · expanded attribution to video-native generative systems.
Unsurface's core technology is Forensic Self-Descriptions (FSD), developed at Drexel University's Multimedia and Information Security Lab (MISL) and peer-reviewed at CVPR 2025 — the top venue in computer vision.
FSD learns the subtle, pixel-level traces that every image creation process leaves behind, without needing prior examples of any specific generator. It was the first method to unify zero-shot detection, open-set source attribution, and forensic clustering of AI-generated imagery in a single framework.
Matthew C. Stamm, CEO, is a Professor of Electrical and Computer Engineering at Drexel University and the founder of MISL, which he established in 2014 to advance the science of multimedia forensics. His lab's research has been supported by DARPA, the Army Research Office, the Defense Forensics Science Center, and the National Science Foundation.
Tai Nguyen, CTO, is the lead author of the FSD paper and a PhD candidate at MISL. He is the primary developer of Unsurface's core detection and attribution technology.
Together, they bring more than a decade of forensic research into operational deployment.