Deepfake Media Detection

Yes, with caveats. Lower resolution reduces the density of forensic signals available, which affects confidence levels. Our reports always document the resolution and quality of the submitted media and their impact on confidence scoring. Even heavily compressed images often retain GAN-specific frequency-domain signatures.

Screenshots from deepfake videos are analyzable as images. Note that screenshot capture itself introduces compression artifacts that partially overlap with manipulation signals - this is documented in our methodology. Where possible, submit the original video rather than a screenshot.

A manipulation heatmap is a visual overlay on the analyzed image or video frame showing the probability of manipulation at each pixel location. Regions with high manipulation probability are highlighted. Heatmaps are generated directly from our forensic analysis and serve as the primary visual evidence in our reports - they are designed to be comprehensible to non-technical judicial officers and can be presented alongside expert testimony.

In some cases, yes. Different generative AI architectures leave characteristic frequency-domain signatures that our analysis can identify. Where a specific model or architecture is identifiable, we document this in the report. In other cases, we can confirm manipulation without being able to identify the specific tool used.

Yes. Social media platforms apply compression when images are uploaded, but the compressed version is what we analyze - and our methodology is calibrated for social-media-compressed imagery. We accept file uploads or URLs for social media content.