Deepfake video is evolving: the most common tells are changing

## Beyond the Uncanny Valley We are leaving the era where deepfakes looked "waxy" or jittery. 2026 models handle lighting changes, profile views, and eye contact with frightening competence. The tell is no longer in the *pixels*, but in the *physics* and the *provenance*. ### The Physics of Light AI struggles to simulate consistent light interaction across complex 3D scenes. * **Reflections**: Does the reflection in the eyes match the light source in the room? * **Shadows**: Do the shadows move perfectly fast enough when the subject moves quickly? ## Why it matters As visual artifacts disappear, verification must move to **provenance** (origin tracking) and **contextual consistency**. ## Quick checks * **Reverse Search**: Does this video exist elsewhere? * **Provenance**: Can you trace the file back to a camera original? (C2PA standards help here). ## What to do next Stay ahead of the curve. 1. Educate your team on the latest Deepfake trends. 2. Use professional [tools to verify image authenticity](/buy-credits) for video keyframes.