How to Detect a Face Swap: 6 Signals to Look For

A face swap can fool you in a thumbnail. It rarely survives a close, structured look. Below are six signals to check on any suspicious photo or video — and a note on why even careful eyes miss the subtle ones.

1. The blending boundary

Every face swap has to stitch a generated or borrowed face onto a real head. The seam — along the jawline, the hairline, and where the cheeks meet the ears — is the weakest point.

Look for a faint halo or colour shift along the edge of the face, hair that appears unnaturally “cut off,” or a jaw that doesn’t quite line up with the neck below it.

2. Skin texture and resolution mismatch

The swapped face is often produced at a different resolution than the body it lands on. The result is a texture mismatch.

Look for a face that is suspiciously smooth — pores and fine lines erased — while the neck, ears and hands keep realistic texture. The opposite also happens: a face that is noticeably sharper than everything around it.

3. Lighting and shadow direction

A real face is lit by the same light sources as the scene behind it. A swapped one frequently isn’t.

Check that the shadows under the nose, chin and brow agree with the light direction in the background. A face that looks “pasted into” the lighting — or catchlights in the eyes that don’t match the scene — is a strong signal.

4. Eyes, gaze and reflections

Eyes are the hardest feature to fake convincingly. Look for the two eyes reflecting different light (catchlights that don’t match each other), pupils that aren’t quite round, a gaze that doesn’t track consistently, or eyes that simply look glassy and lifeless. In video, irregular or absent blinking is a classic tell.

5. Teeth, ears and accessories

Image generators struggle with fine, high-frequency detail. Teeth often render as a single white shape with no individual definition. Ears, earrings and glasses frames may warp, bend, or partially dissolve where they cross the swap boundary.

6. Temporal flicker (video only)

In video, watch the face across frames, not in a single still. Face swaps tend to jitter along the boundary, “pop” or warp during fast head movement, and lose tracking at extreme angles. Lips that drift out of sync with the audio are another red flag. This is also a useful check for deepfakes in live video calls.

Why the eye test isn’t enough

Modern face-swap models are trained to beat exactly these checks at a glance. The signals above are real — but they live in pixel-level detail your eye smooths over, especially after social-media re-compression flattens texture and colour. A still that looks clean to you can still carry forensic traces. That is where automated analysis comes in.

How FakeRadar detects face swaps

FakeRadar runs a dedicated face-swap detector that locates every face in an image, then scores each one independently for manipulation signals — shown as numbered boxes, so you can see exactly which face was flagged and how confident the model is. It layers this on top of ELA, frequency analysis, C2PA credentials and EXIF metadata (see how to read ELA heatmaps and why detection is signal-based).

Crucially, FakeRadar reports signals, not verdicts. No single cue proves a swap on its own; the value is in how the signals stack up across multiple independent layers.

Summary

Treat each of these as one piece of evidence, not a pass/fail test. Read them together, and read them with caution.

Suspect a swapped face? Run it through FakeRadar’s face-swap detector — per-face analysis, free to start.