One video frame cannot authenticate the whole video.

Choose a frame for a specific question

For a location question, choose a clear view of landmarks. For a suspected inserted object, collect surrounding frames as context. For an alleged spoken quote, a face image cannot resolve the central question; you need the full recording and transcript.

Record how the frame was captured. A screen capture can include player controls, subtitles or scaling, while an exported frame may have different dimensions. Those differences matter when comparing files.

Maintain the link to the source

Save the original video URL, visible publication date, timestamp and the caption you are investigating. Name the derivative descriptively without treating that filename as evidence of capture time. Keep any crop as a separate file.

1. Record the video source and exact timestamp before extracting anything.
2. Save a full frame and, if useful, a clearly labeled crop.
3. Inspect the still for supported records or use it to search for matching scenes.
4. Report the result only for that still and explain what the rest of the video still requires.

Avoid the frame-to-video leap

A file-inspection result for a PNG exported from a video belongs to the PNG. Its metadata can describe the export software rather than the camera or production process of the source. A detector estimate for that still is also not an authentication of the voice.

An appropriate report says: “This frame matches an earlier image” or “This exported frame contains these saved records.” It should not turn those observations into an unsupported conclusion that every second of the video is genuine or generated.