Face boundary and blending
Look at the jawline, hairline and ears at full zoom. Face swaps leave a soft or shifting seam, and the swapped region often has a different noise floor from the neck and background.
Guide
Deepfake detection, in practice
Face swaps, re-enactment and cloned voices are now cheap to produce and widely used in scams, harassment and disinformation. This page covers what those techniques leave behind, how to inspect a file yourself, and where automated detection helps or falls short.
Run a deepfake check 3 free scans every day
Six tells to check manually
Eyes, teeth and tongue
Generated interiors of the mouth are a persistent weakness: teeth that merge or change count, a tongue that appears and vanishes, and reflections in the eyes that do not match the room.
Lighting continuity
The face should be lit by the same key light as the shoulders and background, with matching contact shadows under the chin and consistent specular highlights as the head turns.
Temporal drift
Step through frames. Earrings, collars, moles and glasses that mutate between frames are strong evidence of synthesis, since a real camera records a stable subject.
Audio and lip sync
Cloned audio often has flat room tone, missing breaths, and phoneme timing that leads or lags the lips by a fraction of a second.
Provenance
Check whether the clip exists anywhere reputable, whether the account posting it is new, and whether metadata or a C2PA credential survives. Absence is not proof, but a clean chain of custody is powerful.
A repeatable verification routine
- 1
Preserve the evidence: download the original file and record the URL, account and time you found it.
- 2
Establish provenance: search for the same clip elsewhere and check whether any credible outlet carries it.
- 3
Inspect manually: zoom on the face boundary, mouth and eyes, then step through frames looking for drift.
- 4
Run automated analysis: upload the file to DetectFake and read the per-signal report, not just the score.
- 5
Decide and document: combine context with the signals, and keep the report if the file may be disputed later.
Where detection is weakest
Reliability drops sharply with compression, so a clip forwarded through several messaging apps is much harder to judge than the original upload. Short clips give fewer frames to compare, filters and denoising imitate generative smoothness, and screen recordings destroy sensor evidence entirely. Equally, an authentic video shot in unusual lighting can legitimately look wrong. Any honest deepfake verdict is probabilistic, which is why DetectFake shows the reasoning and leaves the final call to you.
Deepfake detection FAQs
What is a deepfake?
A deepfake is media where a real person's face, voice or body has been replaced or synthesized by a machine-learning model. In practice this covers face swaps onto an existing video, face re-enactment that puppets a real person's expressions, fully generated people who never existed, and audio clones used over real footage.
How can you detect a deepfake?
Combine three approaches. First, context: where did the file come from, does any credible source carry the same clip, and does the claim make sense. Second, visual forensics: boundary artifacts at the jaw and hairline, lighting on the face that disagrees with the scene, teeth and eye detail that breaks under motion, and detail that drifts between frames. Third, automated analysis such as DetectFake, which enumerates those signals and returns a confidence score.
Can deepfake detection be fooled?
Yes. Compression from messaging apps and social platforms destroys the fine detail detectors depend on, and re-encoding, beauty filters, screen recordings and very short clips all reduce reliability. Newer generators also improve quickly. This is why a verdict should be treated as evidence for a human decision, not a final ruling.
Is deepfake detection free?
DetectFake gives three free scans per day with the full forensic report — verdict, confidence and every signal. Silver adds 30 scans a day with deeper video frame sampling and Gold is unlimited.
What should I do if I find a deepfake of myself?
Save the original file and the URL with a timestamp before it disappears, run and export a detection report, report it to the hosting platform under its synthetic-media or non-consensual imagery policy, and — if it involves fraud, extortion or intimate imagery — report it to your local law-enforcement cybercrime unit. Do not pay anyone demanding money over it.