01
Upload Target Video
02
How It Works
// DeepTrace forensic pipeline — 3 stage detection
Frame Extraction
Haar cascade face detector samples up to 30 frames with visible faces from the uploaded video.
Neural Analysis
CrossEfficientViT runs each face crop through a sigmoid classifier trained on REAL and FAKE labels.
Verdict
Frame probabilities are averaged. REAL ≥ 65% or FAKE ≥ 60% gives a verdict, else UNCERTAIN.