How to Spot a Deepfake Video
By Christopher Elley, Founder, FactHeck · Last reviewed 28 May 2026
Written with AI assistance and reviewed for accuracy by the author.
Spotting a deepfake video starts with examining the face closely: look for unnatural blinking, blurred or flickering edges around the hairline and ears, and mismatched skin tone between the face and neck. Watch the lighting: shadows that don't follow the face and reflections that look inconsistent are common artefacts. Listen for audio that doesn't quite sync with the lips, or a voice that sounds flat and lacks natural breath. Free tools such as InVID/WeVerify and Hive Moderation can give a second opinion. No single signal is conclusive. Combining a close visual check with a detection tool gives the strongest result.
What is a deepfake, and why it matters
A deepfake is video or audio in which AI has swapped a face or cloned a voice to make someone appear to say or do something they never did. It is different from a simple edit, cut, or filter: the manipulation is generated by a model trained on real footage of the target. Deepfakes spread fastest when they are emotionally charged and feature a public figure, because that combination is highly shareable and low-friction to repost. Researchers at the MIT Media Lab Detect Fakes project and the Alan Turing Institute have documented both how convincing they have become and how to interrogate them.
Visual warning signs
Pause on a clear frame of the face and look closely for:
| Artefact | Where to look | Reliable on its own? |
|---|---|---|
| Blurred / flickering edges | Hairline, ears, jaw, where face meets neck | Fairly |
| Blink / eye anomalies | Blink rate, missing catchlight, blank stare | Weak (newer models fixed this) |
| Lighting mismatch | Shadows on face vs. background | Fairly |
| Skin texture | Waxy or hyper-smooth in close-up | Weak |
| Teeth / mouth | Asymmetric or smeared teeth when talking | Weak |
Audio warning signs
- Lip-sync drift: mouth movements lead or lag the soundtrack.
- Flat prosody: robotic cadence, missing breaths, unnatural pauses.
- Mismatched ambience: background sound that doesn't fit the scene.
Try muting the clip: if the movement still looks natural without the audio cue, that's reassuring; if it falls apart, be suspicious.
Free tools to check a suspected deepfake
| Tool | What it checks | Cost |
|---|---|---|
| InVID/WeVerify | Frame-by-frame view + reverse image search + metadata | Free |
| Hive Moderation | AI-generated-media detector | Free tier |
| FotoForensics | Error-level analysis on extracted still frames | Free |
Run at least two and compare; agreement between independent tools is far more telling than any single score. Microsoft demonstrated a "Video Authenticator" that scores manipulated regions, but it was only ever released to a limited set of organisations rather than as a public tool, so don't count on having access to it.
A step-by-step verification workflow
- Pause on a clear face frame and zoom in on the hairline and ears.
- Mute and watch. Does the movement feel natural without the audio?
- Run the video or an extracted frame through InVID/WeVerify.
- Cross-check with Hive Moderation or another AI-media detector.
- Search for the original clip. Has a credible outlet reported it?
- Check the posting account's history and when it was created.
If you think a video is a deepfake
Don't share it before you've verified it. Report it through the platform's reporting flow. If it depicts a real person, remember they may be a victim. In the UK, sharing non-consensual deepfake intimate imagery is an offence under the Online Safety Act 2023.
The limits of deepfake detection
Be honest about the caveats. Detection accuracy drops sharply once a video has been re-compressed by messaging apps such as WhatsApp or Telegram, and each new generation of models closes more of the visual gaps the checklist above relies on. A clean detector result is reassurance, not proof; a flagged result is a reason to dig deeper, not a verdict. The strongest defence remains tracing the clip back to its original, credible source.
Frequently asked questions
Can you spot a deepfake just by watching the video?
Sometimes. Look for unnatural blinking, a blurred or flickering hairline, mismatched skin tone between face and neck, lighting that doesn't match the background, and audio that drifts out of sync with the lips. No single cue is conclusive; combine a close visual check with a detection tool.
What are the best free tools to detect a deepfake?
InVID/WeVerify (frame-level analysis and reverse image search), Hive Moderation's AI-media detector, and FotoForensics for still frames are all free or have free tiers. Use at least two and compare. Microsoft demonstrated a "Video Authenticator" that scores manipulated regions, but it was only released to a limited set of organisations rather than as a public tool.
Are deepfake detectors always accurate?
No. Detection accuracy drops when a video has been heavily re-compressed (for example after being forwarded on WhatsApp or Telegram), and newer generation models close many of the visual gaps detectors rely on. Treat any result as a probability, not proof, and corroborate with the original source.
What should I do if I think a video is a deepfake?
Don't share it before you've verified it. Check whether a credible outlet has reported the clip, look at the posting account's history, and report it to the platform. If it depicts a real person, they may be a victim; handle it with care.
Want a second opinion in seconds? Paste the video link into FactHeck for an automated AI-detection scan alongside a full claim-by-claim fact-check.