On Camera · 06 filed
Videos and photographs that claim to show a ghost, looked at frame by frame: what the tape shows, which artifact explains it, and how AI ghosts give themselves away.
The first ghost photograph was a double exposure. In the early 1860s a Boston engraver named William Mumler found a faint figure standing behind his own portrait, said it resembled a dead cousin, and within a decade grieving people were paying him for pictures of their dead. The camera has been seeing ghosts ever since: in the tracking noise of a videotape, in dust lit by a flash, in a security camera that records ten frames a second, and now in video that no camera recorded at all.
This section stands at the monitor. Each piece takes one clip, one photograph or one kind of artifact and looks at it closely: what is actually in the frame, which ordinary process could put it there, and what would have to be true for the ordinary explanation to fail. The site does not decide what a ghost is. It says what it can see.
Lead storyImage models paint texture without geometry. The hands, text, reflections, railings and fake aging that give an AI ghost photo away, and the checks to run on one.
More in this section
Plain definitions of the words used around ghost footage and investigations, from apparition and EVP to orbs, rods, REM pods and spirit boxes, with ordinary causes.
Multipath ghosting, VHS noise, macroblocking, rolling shutter, infrared orbs, HDR ghosting: what each artifact looks like on screen and how to check for it.
Text-to-video models draw every frame of a ghost clip at once, from noise. Where that leaves tells, which labels to check, and how to examine a clip frame by frame.
What orbs, mist, faces and transparent figures in ghost photos usually are: backscatter, breath, pareidolia, long exposure, and how to test your own pictures.