Reports

Case studies

Real-world examples of ClipForensics forensic analysis applied to deepfakes, viral videos, and corporate compliance scenarios.

Deepfake Detection

Detecting a face-swap deepfake in a political video

A viral video purporting to show a politician making controversial statements was submitted for analysis. ClipForensics identified face-swap artifacts, compression chain anomalies, and lip-sync misalignment.

Key Modules

Face Manipulation, Lip Sync, Compression History, Metadata

Verdict

High Manipulation Risk

Analysis Time

< 60 seconds

Key Finding

The face manipulation module detected boundary artifacts at the jawline. The compression history showed three encoding passes — one from the original camera, one from an editing tool, and one from social media upload. The lip-sync module found statistically significant phoneme-viseme misalignment in multiple segments.

Authenticity Verification

Verifying citizen journalism footage from a conflict zone

A news organization received smartphone footage claiming to document a specific event. They needed to verify the footage was authentic before publication.

Key Modules

Metadata, Provenance, Compression, Temporal Consistency, Lighting

Verdict

Likely Authentic

Analysis Time

< 60 seconds

Key Finding

Metadata was consistent with the claimed device (iPhone 15 Pro). Single-pass encoding from Apple VideoToolbox. Temporal consistency, lighting, and noise patterns were all consistent with authentic camera footage. No face manipulation or AI generation signals detected.

Corporate Compliance

Identifying AI-generated product testimonial videos

A brand protection team discovered product review videos on social media featuring people who didn't appear in any other online context. They submitted the videos for forensic analysis.

Key Modules

Spectral Analysis, Biological Motion, Optical Flow, Metadata, Audio Synthesis

Verdict

Likely AI-Generated

Analysis Time

< 60 seconds

Key Finding

Spectral analysis detected frequency patterns consistent with diffusion-model generation. Biological motion scoring showed subtle gait anomalies. Metadata contained no camera information. Audio showed spectral patterns consistent with TTS synthesis.

Analyze your own videos

See the same forensic analysis applied to your content.