Video fakes split into two distinct problems: fully AI-generated video (Sora, Runway, Kling) that creates synthetic footage from scratch, and deepfake video that swaps one real person's face into another real video. Detection approaches differ significantly between them. Here's how each works, which consumer tools cover video in 2026, and where FauxSpy fits in the stack.
AI-generated video
Created entirely from a text prompt or reference image. The camera never captured any of this footage. Examples: Sora, Runway, Pika, Kling, Emu Video.
Detection looks for: synthetic generation artifacts across all pixels, unnatural motion physics, temporal frequency artifacts in the whole frame
Deepfake video
Real camera footage with one or more faces swapped using AI. The background, body, and scene are real; the face is replaced. Examples: FaceSwap, DeepFaceLab, commercial APIs.
Detection looks for: frame-to-frame face inconsistencies, boundary blending artifacts, biometric inconsistencies across frames
Most consumer deepfake concerns — a video call that seems off, a viral video of a public figure saying something suspicious, a video sent in a romance scam — involve face-swap deepfakes rather than fully synthetic AI video. Fully AI-generated video is still visually distinguishable to trained eyes (unnatural motion, physics violations, texture anomalies) at current quality levels, though this is changing rapidly.
Unlike still-image detection, video deepfake detection has an additional signal source: temporal coherence across frames. A face-swap algorithm processes each frame with some independence — the blending parameters shift slightly, the skin texture varies slightly frame to frame, the positioning of the swapped face may drift. In a real video, the same face has a consistent statistical signature across frames. In a deepfake, the swapped face region shows frame-to-frame variation patterns that differ from how real faces behave in video.
Temporal inconsistency analysis
Detector compares the face region across frames, looking for variation patterns inconsistent with natural face movement. A swapped face "flickers" slightly because the face-swap algorithm isn't perfectly temporally coherent.
Face boundary artifacts
The boundary between the swapped face and the original neck and jaw has characteristic blending artifacts — frequency patterns at the transition that differ from how real skin-to-skin boundaries appear in camera footage. Detectors segment the face and analyze boundary statistics.
Biometric inconsistency
Eye gaze, pupil reflections, and skin light response should be consistent with the scene's light source. Deepfakes often produce subtle inconsistencies in how the swapped face responds to lighting across frames, or micro-expression timing that doesn't match the body language.
| Tool | Video deepfake | AI-gen video | Free tier | Extension |
|---|---|---|---|---|
| FauxSpy Pro + Video | ✓ | ✓ | $29.99/mo | ✓ Chrome + Firefox |
| BitMind | ✓ | Partial | Free, no account | ✓ Chrome + Edge |
| Resemble AI | ✓ | ✓ | 4/day free | ✓ Chrome |
| UncovAI | ✓ | Partial | Free tier | ✓ Chrome |
| Hive Moderation (API) | ✓ | ✓ | No free tier | No extension |
Table reflects mid-2026 research. Verify current features on each tool's site before purchasing.
FauxSpy's primary focus is still-image AI detection — the use case it handles best, with no account required and 3 free checks per day. The Pro tier ($9.99/month) adds Sightengine's deepfake model for still-image face-swap detection. The Pro + Video tier ($29.99/month) extends this to video content analysis.
For users who primarily check still images (profile photos, product listings, news images) and occasionally need video checking, Pro + Video covers both. For users who only need video deepfake detection and don't have a still-image use case, BitMind's free tier (anonymous, no account, video included) may be more cost-effective.
See the deepfake detector page for a complete breakdown of FauxSpy's deepfake-specific capabilities.
AI-generated video (Sora, Runway, Kling) creates entirely synthetic footage from a text prompt — no real camera footage involved. A deepfake uses real video with a specific face replaced by AI. Detection differs: AI-generated video is caught by analyzing synthetic artifacts across all frames; deepfakes are caught by looking for face boundary artifacts and frame-to-frame temporal inconsistencies at the face region specifically.
Three main signals: temporal inconsistency (face-swap algorithms process frames somewhat independently, creating subtle frame-to-frame flickering); face boundary artifacts (the blending region between swapped face and original has characteristic statistical differences); biometric inconsistencies (eye gaze, skin reflections, micro-expressions may be inconsistent with the scene across frames). All three are more reliable signals than any single still-frame check.
FauxSpy Pro + Video ($29.99/month) includes video checking using Sightengine's deepfake model. The free tier and standard Pro check still images only. For free video deepfake checking, BitMind (free, no account, Chrome/Edge extension) includes video analysis. Resemble AI offers 4 free video checks per day via its Chrome extension.
Yes. BitMind (Chrome/Edge, free, no account) works on video content in-browser. Resemble AI's Chrome extension covers video with 4 free checks per day. UncovAI also includes video. FauxSpy Pro + Video adds this to the FauxSpy extension. For a quick first pass on a social media video, right-clicking a suspicious frame and running a still-image check with FauxSpy free is a reasonable starting point before running full video analysis.
3 checks/day for AI-generated images. Upgrade to Pro + Video when you need video deepfake analysis.
🕵️ Add to Chrome — Free See Pro + Video