← Play Real or AI?

How to spot AI-generated videos

There is no single magic tell. The best approach is to combine several visual and contextual clues, then verify the source before making a confident claim.

Start with context, not pixels. A strange hand can appear in real compression artifacts, and a flawless clip can still be synthetic. Treat every clue as evidence—not proof by itself.

Eight checks that work together

  1. Check who posted it first. Look for an original upload, a named creator, a reliable publication, or another traceable source. Reposts with cropped captions and no provenance deserve more scrutiny.
  2. Watch the same object across several frames. Pause and scrub. AI video may let jewelry, fingers, tools, patterns, or background objects change shape, swap sides, merge, or disappear as the camera moves.
  3. Look at contact and physics. Pay attention when feet meet the floor, hands grip objects, fabric touches bodies, liquid splashes, or objects collide. Sliding contact points and weightless motion are stronger clues than one imperfect still frame.
  4. Read small text and repeated symbols. Signs, labels, jersey numbers, interfaces, and logos may mutate between frames. Modern models can render good text, so consistency over time matters more than a single readable word.
  5. Compare light, shadows, and reflections. Ask whether every subject appears to share the same light source. Reflections that omit objects, shadows that move independently, or highlights that ignore the scene can reveal synthesis.
  6. Listen separately from watching. Check whether speech, breaths, impacts, and room sound match what happens on screen. Lip movement that drifts from consonants or ambience that never changes can be a warning sign.
  7. Inspect edges during fast motion. Hair, glasses, spokes, fingers, and thin objects can melt into nearby surfaces for a few frames. Slow playback is more useful here than zooming into a static screenshot.
  8. Corroborate before you share. Search for other angles or reputable coverage, extract a key frame for reverse-image search, and check for available provenance or Content Credentials. Missing metadata is not proof of AI, but verified provenance can strengthen confidence.

Common false alarms

Low bitrate, frame interpolation, stabilization, portrait-mode blur, filters, game footage, and aggressive social-media recompression can all look “AI-ish.” That is why a verdict based only on odd fingers, smooth skin, or one warped frame is unreliable.

A simple decision rule

Use three layers: source (where did it come from?), time (do details stay consistent across frames?), and physics (do motion, contact, light, and sound agree?). If only one layer looks suspicious, keep investigating. If several independent clues fail, confidence can rise—but label uncertainty honestly.

Test your eye

Try five real and AI photos or videos, get every answer immediately, and compare your calls with the crowd.