What Happens When You Upload a Video?
When you upload a video, DeepScan AI doesn't just look at one frame — it samples multiple frames from across the entire video, finds faces in each frame, and runs five different checks on them. Each check looks for a different type of manipulation. The results are combined into a single score that determines whether the video is REAL or FAKE.
The whole process takes between 10 and 40 seconds depending on video length and how many frames you choose to analyze.
Step 1 — Frame Sampling
A video is made up of hundreds or thousands of individual images (frames). Analyzing every single frame would take too long, so DeepScan AI picks a set number of frames spread evenly from the beginning to the end of the video. You can choose how many frames to analyze — more frames means higher accuracy but takes longer.
Step 2 — Face Detection
In each sampled frame, the system automatically finds and crops the face region. This is important because deepfake manipulation almost always targets the face. By focusing on the face, the analysis is more precise and less affected by background noise. If no face is found in a frame, the center of the image is used instead.
Step 3 — Five Detection Checks
Five independent checks run on the face crops. Each one looks for a different sign of manipulation.
🧠 AI Feature Consistency
A deep learning model analyzes each face and creates a unique "fingerprint" for it. In real videos, these fingerprints are very similar from frame to frame. In deepfakes, they jump around erratically because the AI regenerates the face each time.
🔬 Sharpness Consistency
Real videos have consistently sharp faces throughout. Deepfake generators sometimes produce blurry frames mixed with sharp ones. This check measures how much the sharpness varies across frames.
✂️ Face Boundary Check
When a face is swapped onto a different body, there's often a visible "seam" at the edge where the face meets the neck and background. This check looks for unusually sharp edges at the face boundary — a telltale sign of face-swap deepfakes.
🎨 Color Matching
In a real video, the face and background are lit by the same light source, so their colors match naturally. In a deepfake, the face is often taken from a different video with different lighting, causing a color mismatch between the face and the rest of the scene.
📈 Motion Smoothness
Real faces move smoothly and naturally between frames. Deepfake generators can produce subtle flickering or unnatural jumps in the face region. This check measures how smoothly the face changes from one frame to the next.
⚖️ Combined Score
The five checks are combined into a single suspicion score. The AI feature check carries the most weight (45%) since it's the most reliable. If the total score crosses a threshold, the video is flagged as FAKE.
Step 4 — Verdict
After all five checks are complete, the scores are combined into a final suspicion index from 0 to 100. A score below 38 means the video shows no significant signs of manipulation — verdict: REAL. A score of 38 or above means multiple manipulation signals were detected — verdict: FAKE.
The confidence percentage tells you how certain the system is. A high confidence REAL result means the video passed all checks comfortably. A high confidence FAKE result means multiple strong signals were detected.
Important to Know
DeepScan AI is a detection tool, not a definitive judge. It works best on face-swap deepfakes — videos where someone's face has been replaced with another person's. It may be less effective on very short clips, very low quality videos, or highly sophisticated AI-generated content where manipulation artifacts are minimal.
Always treat results as one piece of evidence, not a final verdict.