This is one of the most common questions about deepfakes. The honest answer: it's getting harder. But there are still things you can look for. This guide covers practical detection methods for regular people, not just forensic experts.
The Short Answer
Look for these red flags:
- Face edges – Blurry or wavy boundaries around the face
- Skin texture – Too smooth, like plastic
- Blinking – Too little, too much, or unnaturally timed
- Lighting – Face lit differently than the rest of the scene
- Audio sync – Lips don't quite match the words
- Context – Does this video make sense? Who benefits from it being real?
None of these are foolproof. High-quality deepfakes may show none of them. But most deepfakes still have at least one tell.
Detailed Detection: What to Look For
Q: What should I look at first?
A: The face boundary.
The edge where the swapped face meets the original head is the hardest part to fake perfectly. Look for:
- Blurring around the jawline or hairline
- Color difference between face and neck
- Waviness or instability at the edges during motion
- A sense that the face is "floating" on top of the head
How to check: Pause the video during head movement. Look closely at where the face meets the hair and neck.
Q: What about the skin?
A: Real skin has texture. Deepfake skin is often too perfect.
Look for:
- Missing pores and fine lines
- Uniform skin that looks airbrushed
- A waxy or plastic appearance
- Skin that doesn't match hands, neck, or ears
How to check: Compare the face to any other visible skin (neck, hands, ears). Real skin has consistent texture everywhere.
Q: Everyone mentions blinking. Does that still work?
A: Yes, but it's not as reliable as it used to be.
Early deepfakes famously forgot to include blinking. Modern systems usually include it, but often get it wrong:
- Too little: Real people blink 15-20 times per minute
- Too regular: Natural blinking is irregular
- Too synchronized: Both eyes blink at exactly the same rate
How to check: Watch 30-60 seconds of video. Count blinks. If they're absent, perfectly regular, or just feel "off," that's suspicious.
Q: What about the eyes themselves?
A: Eyes are hard to fake well.
Look for:
- Dead stare: Eyes that look empty or lifeless
- Wrong reflections: Shapes reflected in the eyes that don't match the scene
- Gaze mismatch: Looking slightly in the wrong direction
- Inconsistent reflections: Left and right eye showing different things
How to check: Zoom in on the eyes. Can you see what's reflected? Does it match what should be in front of the person?
Q: How do I check the lighting?
A: Shadows and highlights should match the scene.
Look for:
- Shadows on the face pointing a different direction than shadows elsewhere
- Face that's brighter or darker than it should be given the lighting
- Shiny spots (specular highlights) that don't correspond to light sources
- Face color temperature that doesn't match the room
How to check: Identify where the light is coming from in the scene. Do the shadows on the face match that direction?
Q: What about audio sync?
A: Even good deepfakes often have slight sync issues.
Look for:
- Lips that don't quite match the words
- Consonants that require lip closure (B, M, P) where lips don't fully close
- Audio that seems slightly ahead or behind the video
- Speech that sounds natural but looks slightly off
How to check: Watch for words like "baby," "mama," "papa." The lips must close completely for these sounds.
Q: Are there things deepfakes can't do well?
A: Yes. Look for these difficult scenarios:
- Teeth: Often blurry or merged together
- Profile views: Side angles are harder to fake
- Rapid motion: Fast head turns often cause glitches
- Hands near face: Usually creates artifacts
- Eating or drinking: Almost always looks wrong
- Strong expressions: Extreme emotions are harder to fake
How to check: Does the video conveniently avoid these scenarios? That might be intentional.
Beyond Visual Detection
Q: Can I use software to detect deepfakes?
A: Yes, but with limitations.
Available tools:
| Tool Type | Accessibility | Accuracy | Limitations |
|---|---|---|---|
| Free online detectors | Easy | 60-80% | Many false positives/negatives |
| Research tools | Requires technical skill | 80-90% | May not work on new deepfake types |
| Commercial services | Paid, easy to use | 85-95% | Expensive, not foolproof |
The reality: No tool is perfect. Detection accuracy drops significantly on:
- Heavily compressed video (social media)
- New deepfake generation methods
- High-quality production deepfakes
Q: Does compression make detection harder?
A: Yes, significantly.
Social media platforms compress video heavily. This removes many of the subtle signals that detection tools rely on. A deepfake that might be detectable at original quality becomes much harder to identify after:
- Instagram compression
- TikTok processing
- Twitter/X video handling
- WhatsApp sharing
- Screenshot and re-upload cycles
The reality: If you're checking video that's been through social media, visual inspection may be more reliable than automated tools.
Q: What about checking the source?
A: This is often more reliable than visual detection.
Ask:
- Where did this video come from? Is it from a verified account?
- Has anyone else confirmed it? Can you find the same video from independent sources?
- Who benefits? If the video is true, who gains from it?
- Does it fit the pattern? Is this person likely to say/do what's shown?
- Is there a response? Has the person in the video addressed it?
The reality: Verification is often more reliable than detection. A video from an unverified source making extraordinary claims should be treated skeptically regardless of whether you can spot visual artifacts.
Common Mistakes in Detection
Q: What mistakes do people make when trying to spot deepfakes?
A: Several common ones:
1. Expecting obvious fakes
"I'd definitely be able to tell."
Most people overestimate their ability. High-quality deepfakes fool most viewers.
2. Trusting single indicators
"The blinking looked normal, so it must be real."
Real videos can have unusual blinking. Deepfakes can have normal blinking. No single indicator is definitive.
3. Ignoring context
"The face looked perfect, so I believed it."
Even if you can't detect visual artifacts, ask: does this make sense? Why would this video exist?
4. Assuming old = safe
"Deepfakes are new, so older videos must be real."
Old videos can be deepfaked now. And some manipulation techniques predate modern deepfakes.
5. Over-relying on tools
"The detector said it's real, so it must be."
Detection tools have false negatives. A "real" result doesn't guarantee authenticity.
When Detection Fails
Q: What if I can't tell whether something is fake?
A: This is increasingly common. Here's what to do:
-
Don't share it immediately. If you're unsure, don't spread it.
-
Wait for verification. Major claims will be investigated. Give it time.
-
Check multiple sources. If only one source has the video, be skeptical.
-
Consider the stakes. High-stakes claims (political, financial, reputational) deserve more scrutiny.
-
Look for official responses. The person in the video may confirm or deny.
-
Accept uncertainty. Sometimes you won't be able to tell. That's okay.
Q: Is detection going to get easier or harder?
A: Harder, at least in the short term.
- Generation technology is improving faster than detection
- New methods appear regularly
- Compression removes detection signals
- The arms race favors generators
Long-term possibilities:
- Content authentication (proving real, not detecting fake)
- Blockchain-based provenance tracking
- Camera-level authentication
- AI-powered real-time detection
But for now, expect detection to remain difficult and imperfect.
Practical Checklist
Before believing a video:
- [ ] Check the source—is it verified and trustworthy?
- [ ] Look for independent confirmation
- [ ] Examine face edges during motion
- [ ] Check skin texture vs. other visible skin
- [ ] Watch blinking patterns
- [ ] Look at eye reflections
- [ ] Verify lighting consistency
- [ ] Test audio sync on consonants
- [ ] Consider: who benefits if this is believed?
- [ ] When in doubt, wait for verification
Summary
Detecting deepfakes is becoming harder as the technology improves. Visual inspection still works for many deepfakes—look for face boundary issues, unnatural skin, blinking problems, lighting mismatches, and audio sync failures. But high-quality deepfakes may show none of these tells.
The most reliable approach combines visual inspection with source verification. Ask not just "does this look real?" but "does this make sense?" and "where did this come from?"
When you can't tell, the right response is skepticism, not assumption. Don't share content you can't verify. Wait for independent confirmation. And accept that perfect detection may not be possible.
Related Topics
- Why Do Deepfakes Still Look Wrong? Common Failure Modes – Visual tells explained
- When Do Deepfakes Break the Laws of Physics? – Physics violations to spot
- Why Does My Deepfake Face Look Wrong? – Facial detail problems
- Does Deepfake Technology Threaten Your Privacy? – Privacy implications
- What Can't Deepfakes Do Yet? – Current technology limits