An experiment with altered images and AI-generated video found that viewers could later remember details that were not in the original material.
Restoring a faded photograph and adding something that was never in it are different jobs, even when an app offers both under the same button. A scratch can be removed from a family portrait. A missing person can also be added, a face changed, or a still image made to move. Only the first photograph records what the camera captured. The distinction becomes harder to keep track of when the edited version is the one everyone saves and shares. A generated smile or movement may become familiar through repetition. Later, someone may remember that detail without remembering that it came from an edit.
A study using altered images and AI-made video tested this kind of memory error. It did not study families restoring their own albums, so we should not claim that result. It does provide a reason to label substantial edits clearly and keep the original files. People who see the image later need a way to tell a record of the past from a new interpretation of it.
Pataranutaporn and his team ran a test with 200 people. It used images and AI-made video. People saw the first set. Later, they saw versions that could have changed details. The changed material led to false memories. The effect was strongest when people saw video made from changed images. The source used here is the 2024 author preprint. It is not presented as a final peer-reviewed version. The study concerns memory for material shown in a set experiment. It does not tell us how all people recall their own family photos after an edit.
The basic warning is clear enough: a later image can be more than decoration. It can change what we use to recall the first one.
Consider an invented photo of a birthday meal. The print has a scratch across the table. Removing the scratch may help you see what was already there. Adding a cake where no cake is visible is a different act. So is making a person turn their head in an animated version. The movement may look natural. It is still a new guess. The camera did not capture that move. An app may describe both changes as restoration or enhancement, which does not tell a viewer whether new content was created. A useful label identifies the actual change, such as adding a background or generating movement from a still image.
Keep the original next to the changed version. A viewer should not have to rely on someone's memory of which file was edited.
An original photograph records a view captured at a particular moment. A generated extension or animation also contains choices made later. Even when those additions look realistic, they should not be treated as information recorded by the camera. Imagine a school photo where a tool fills in the edge of the frame. It adds a window. Years later, someone points to it as proof of which classroom the group used. A visual guess has become a claim about a place.
This example is invented, but the rule is practical. Before using an image as evidence, ask which parts were captured and which were created. If you do not know which is which, say so. The same care applies to captions. A real photo with the wrong date can mislead without changing a single pixel.
A note buried in a folder is not much help when an image gets shared alone. Put a clear label beside the image. Say what changed: AI added the background. Or: Made to move from a still photo. Or: Color added later. Use words that name the change. Enhanced says very little. It could mean brighter light or an invented face. People should not have to guess which. For family photos, keep a short note. List the first date, the edit date, and what changed. You do not need a long technical report. You need enough for the next person to tell a record from a new piece of art.
Do not delete the old file because the new one looks nicer. The old file may hold the very detail a future reader needs.
Suppose you have watched an animated family portrait many times. You can picture the smile and the turn of the head. That tells you the clip is easy to bring to mind. It does not prove the motion took place. If someone asks what you remember, separate the sources. I remember the day. I have also seen an animation of the photo. I am not sure whether that gesture is from the day or the clip. That may sound less tidy than one clear story. It is more useful. It lets other people know where your confidence comes from.
This does not require treating every memory as unreliable. It requires identifying the source of a specific detail when an original photograph and a generated version show different things.
There is nothing inherently wrong with making an imagined version of an old photograph. The problem arises when the edited version is treated as evidence of what happened. Keep the original, explain what was added, and make sure that explanation travels with the image when it is shared. An animation can be emotionally powerful without being a recording. A restored image can be useful without every reconstructed detail being trustworthy. Those distinctions are easy to preserve at the time of editing and much harder to recover years later, when nobody remembers which version came first.
