Deepfake videos are getting harder to catch. The generation of tools available in 2026 produces output that fools most people at normal playback speed. But the technology still leaves traces, and if you know where to look, you can spot them.
These eight red flags are ordered from easiest to check to hardest. You don't need special software for any of them. A phone screen, a pause button, and a bit of patience will do.
The 8 Red Flags
Lip Sync Is Slightly Off
This is the first thing to check. Watch the mouth closely during hard consonant sounds like B, P, and M. These sounds require the lips to fully close. In a deepfake, the lips often stay slightly apart on these sounds, or they close a fraction of a second too late. The mismatch is subtle at normal speed but obvious when you slow the video to 0.5x or 0.25x playback. Pay extra attention when the person says words that start with "B" or "P" -- the lips should press together before the sound comes out.
Weird Earlobes or Jewellery
AI struggles with small, irregular details. Earlobes are a consistent weak point -- they often look melted, asymmetric, or change shape between frames. Earrings are even worse. Watch for earrings that flicker, disappear for a frame, or change shape when the head turns. Glasses frames sometimes warp near the temples. If someone is wearing a necklace, check where it meets the neck -- deepfakes frequently glitch at that boundary.
Inconsistent Lighting on Face vs Background
Real light comes from specific directions. The shadows on someone's face should match the light source visible in the background. In a deepfake, the face is generated separately from the body and background, so the lighting often comes from a different angle. Look at the shadows under the nose and chin. If the background suggests light is coming from the left but the nose shadow falls the wrong way, that's a strong indicator. Skin tone shifts when the head turns are another giveaway -- the AI sometimes renders different colour temperatures for different angles of the same face.
Blurring Around the Jawline and Hairline
This is the blending zone -- where the AI-generated face meets the original footage. Pause the video and look carefully at the edges of the face. The jawline often shows a faint smudge or blur that's not present on the rest of the skin. The hairline is particularly telling. Real hair has individual strands that overlap the forehead in irregular patterns. Deepfake hair tends to look painted on, with a sharp or oddly smooth transition where hair meets skin. Sideburns and facial hair along the jaw are especially problematic for the AI.
Eyes That Don't Blink Naturally
Humans blink roughly 15 to 20 times per minute, with irregular timing. Early deepfakes barely blinked at all. Current models have fixed that obvious tell, but they've overcorrected -- many deepfakes now blink at suspiciously regular intervals, like a metronome. Watch 30 seconds of footage and count the blinks. If they come at perfectly even intervals, that's a flag. Also watch the eyelids themselves. Real blinks involve the upper lid moving down smoothly. Deepfake blinks sometimes show the eye snapping shut and open rather than moving through a natural arc.
Teeth That Look Flat or Blurred
Open the video full-screen and wait for the person to smile or speak with their mouth open. Real teeth have individual shapes, slight gaps, and catch light at different angles. Deepfake teeth often look like a flat white strip, or they blur together into an indistinct mass. Individual teeth may lack definition or shift position between frames. The tongue is another problem area -- it sometimes clips through the teeth or appears to be the wrong size for the mouth.
Audio Quality Mismatch
Listen with headphones. In a real video, the voice and the background audio were recorded by the same microphone in the same room. They share the same reverb, the same ambient noise floor, the same compression artefacts. In a deepfake with cloned audio, the voice often sounds like it was recorded in a different space -- cleaner or more processed than the background. The voice might sound slightly robotic on certain vowel sounds, or there might be tiny gaps where the AI stitched audio segments together. If the voice sounds studio-quality but the video looks like it was shot on a phone, something is off.
Hands and Fingers
AI still struggles badly with hands. If the person in the video gestures or holds something, look at their fingers. Common problems include the wrong number of fingers, fingers that bend at unnatural angles, fingers that merge together, and thumbs that appear on the wrong side of the hand. Hands that pass in front of the face often cause the whole face to glitch for a frame or two. This is one of the most reliable tells -- if the hands look wrong, treat the entire video with suspicion.
How to Slow Down and Check
Most of these red flags are invisible at normal playback speed. You need to slow down.
On YouTube: Click the gear icon, select "Playback speed", and choose 0.25. This makes lip sync errors, blinking patterns, and frame-level artefacts obvious.
On your phone: Screen-record the video if you can't control playback speed directly. Most phones let you scrub through a recorded video frame by frame.
On any platform: Use the spacebar to pause, then use the comma and period keys (on desktop) to step forward and backward one frame at a time. This works on YouTube, Vimeo, and most web-based players.
The Screenshot and Zoom Method
When something looks off but you can't pinpoint it, take a screenshot and zoom in. Pinch-to-zoom on your phone or use your operating system's zoom tool. Focus on these areas:
- The boundary between face and hair
- Earlobes and any jewellery
- The line where the chin meets the neck
- Teeth and the inside of the mouth
- Fingers and fingernails
At 200% to 300% zoom, blending artefacts that are invisible at normal size become obvious. You'll see smudging, colour shifts, or unnatural smoothness at the boundaries where the AI-generated content meets the original footage.
No single red flag proves a video is fake. But three or more red flags in the same video should make you deeply sceptical. When in doubt, don't share it. Check the source, reverse-image search key frames, and report suspicious content to the platform.
Deepfake technology will keep improving. The red flags listed here are accurate as of early 2026, but some will be fixed by next year's models. The best defence isn't any single technique -- it's the habit of pausing before you trust, slowing down before you share, and asking "where did this come from?" before you react.