Video Forensics & Deepfake Analysis
Free AI Video & Deepfake Frame Inspector
Inspect suspicious video clips frame-by-frame right in your browser. Spot unnatural facial warping, deepfake face-swaps, and telltale AI motion flickering — with zero video files uploaded to our servers.
Drag and drop a video to extract forensic frames
Decodes video timeline samples in browser memory to inspect deepfake anomalies, mouth warping, and temporal jitter. Supports MP4, WebM, and MOV up to 60 MB.
Temporal Forensic Framework
Frame-by-Frame Video Inspection Timeline
AI video generators often struggle to maintain physical consistency across consecutive frames. Inspect adjacent frames for these key visual indicators:
Baseline Reference Frame
Establish facial symmetry, iris catchlight reflections, hair strand definition, and skin micro-textures before camera movement begins.
Head Turn & Jawline Boundary
Evaluate rotation angles for double jawlines, soft edge feathering, or ghost blending masks along the collar and neckline.
Biometric Stability
Examine teeth alignment, earlobe contours, and glasses frames to confirm they do not warp, duplicate, or drift between frames.
Texture & Pattern Coherence
Confirm high-frequency textures (brick walls, mesh fabric, leaf foliage) do not simmer, melt, or boil during camera translation.
The 4-Vector Forensic Diagnostic Framework for Video Media
Professional video forensic analysts evaluate synthetic media across four distinct physical and anatomical dimensions:
Temporal Coherence
Diffusion models generate video in localized latent blocks. Inspect fine background details — like patterned wallpaper, leaves, or mesh fabric — for "swimming" textures that boil or morph during camera pans.
Biometric Invariance
Natural human eyes exhibit involuntary micro-saccades and blink completely every 3 to 6 seconds. Deepfakes frequently feature incomplete blinks, asymmetric pupil catchlights, or static gaze vectors.
Boundary Blending
Face-swap networks (RoOP, DeepFaceLive) swap masks over a base actor. Fast head turns reveal blending halos along the jawline, ghost collars, or earlobes momentarily disappearing behind hair.
Phoneme Sync
Compare speech audio tracks against viseme shapes. Plosive sounds (B, P, M) physically require complete lip closure. Generative lip-sync pipelines frequently produce sound with open mouths.
Generative Video Engine Artifact Profiles (2025/2026 Models)
Different generative video architectures leave characteristic digital footprints:
| Generative Engine / Architecture | Dominant Forensic Anomaly | Detection Reliability |
|---|---|---|
| OpenAI Sora (Diffusion Transformer) | Hyper-realistic surface rendering, but subtle physics violations during complex object collisions or liquid pouring. | High via timeline motion vector cross-correlation. |
| Runway Gen-3 Alpha | Excessive directional motion blur during camera pans; high-frequency texture smearing on textile fabrics. | Very High via frame-rate frequency spectrum analysis. |
| Kling AI / MiniMax | High fidelity on Asian facial landmarks, but persistent plastic over-smoothing on skin pores during low-light scenes. | High via localized sensor noise extraction. |
| Luma Dream Machine | Topology morphing — background vehicles or architecture subtly changing shape during 360-degree rotations. | Very High via 3D epipolar geometry reconstruction. |
| Face-Swap / Deepfake Mask | Color temperature mismatch between facial mask and neck; teeth boundary blur during rapid continuous speech. | Extremely High via jawline gradient edge filtering. |
Frequently Asked Questions
Can an AI video or Deepfake be detected entirely in a web browser?
Yes, preliminary triage can be performed locally. Modern web browsers can decode video containers in GPU memory and extract chronological keyframes. Examining adjacent frames reveals temporal inconsistencies, facial boundary flickering, and warping artifacts that are characteristic of generative video models.
What is temporal jitter in generative AI video?
Generative video diffusion models synthesize video frame-by-frame or in short latent blocks. Because these models lack physical awareness of 3D object permanence, fine details (like tooth structure, jewelry, eye reflections, and background textures) constantly fluctuate or shimmer between frames.
How do face-swap deepfakes differ from full generative AI video?
Face-swap deepfakes overlay a synthetic face mask onto an authentic human actor's body. They typically exhibit telltale boundary blending blur along the jawline, unnatural lighting angles relative to the collarbone, and mismatched eye-blink intervals. Full generative videos (e.g. Sora, Runway, Kling) show physics violations across the entire scene.
Does this tool upload my video to a remote server?
No. This tool decodes your video locally using the HTML5 media subsystem in client memory. Your original video file never leaves your computer.