Gym Transformation Photos: The Science of How They Rewrite What Your Members Actually Remember

Every gym that posts a transformation photo is betting that the image shows what actually happened. A 2025 MIT study suggests the image might be replacing what actually happened.

In 1974, Elizabeth Loftus showed research participants a film of a car accident, then asked them how fast the vehicles were going when they “smashed” into each other. A different group got the same film but the word “contacted.” The “smashed” group remembered the cars going faster, and they remembered broken glass scattered across the road. There was no broken glass in the film.

Fifty years of research followed. Loftus became the most cited female psychologist of the twentieth century, largely by proving one uncomfortable idea over and over: memory is a reconstruction, not a recording, and reconstructions can be edited by anyone with the right input at the right moment.

In April 2025, Loftus put her name on a study that should unsettle anyone using gym transformation photos to sell memberships.

Pat Pataranutaporn and a team at MIT Media Lab recruited 200 participants through CloudResearch. All US residents, aged 20 to 73, split evenly by gender. The study was pre-registered and went on to win Best Paper at CHI, the world’s top human-computer interaction conference.

Participants viewed 24 original photographs in a swipeable, Instagram-style interface for two minutes, the way you would scroll a feed on your phone. Then came a filler task (Pac-Man, of all things) to create a time gap before a second viewing, which is where the design did its work.

The researchers split 200 people into four groups of fifty. The control group saw the same unedited images a second time, while the second group saw AI-edited versions of those photographs. A third group saw AI-generated videos, and the fourth saw AI-generated videos made from the AI-edited images, the full stack of edit the photo then animate it.

Every altered image carried the label “AI-enhanced image.” Participants were told, explicitly, that what they were seeing had been modified by artificial intelligence. It made no measurable difference to what happened next.

The fourth group, participants who saw AI-generated videos of AI-edited images, formed false memories at 2.05 times the rate of the control group, which is not a marginal uplift but a doubling of the false recollection rate. Their confidence in those false memories was 1.19 times higher than control as well.

Knowing the images were AI-enhanced did not protect against the memory distortion. The visual input overwrote what participants had actually seen, and they reported their altered recollections as memories they trusted.

The researchers were not trying to trick anyone, and the images were not pretending to be real. Participants had full disclosure about every modification. The mechanism worked regardless.

Why labels fail

On the surface, you would expect disclosure to solve this. If someone knows an image has been edited, they should adjust for it, discount the alteration, treat the visual as unreliable.

Visual memory does not cooperate with that logic.

When you see a vivid image, your brain processes it through the same neural pathways regardless of what a caption underneath it says. Your visual cortex does not read disclaimers before it encodes what it sees, and when the time comes to recall, it draws on whatever encoding is strongest. An AI-enhanced image, sharper and more vivid than the original, often wins that competition simply because it left a deeper perceptual trace.

Loftus’s misinformation effect has documented this process for decades. Post-event information integrates into the existing memory and becomes indistinguishable from the original experience. The MIT study extends the principle into AI-generated media, where the fidelity of the altered version can exceed what the person actually saw in the first place.

The gradient across the four conditions tells its own story. AI-edited still images increased false memories compared to the control. AI-generated videos increased them further. But the combination of AI-edited images turned into AI-generated videos produced the strongest effect of all, because more visual fidelity meant more memory distortion. The richer the fabrication, the deeper it embedded itself.

The transformation photo economy

The fitness industry runs on visual transformation, and before-and-after photos remain the single most persuasive marketing asset most gyms possess. Member testimonials, challenge results, and progress documentation all rest on the assumption that a photograph records what actually happened.

Consider what passes as standard practice already: lighting adjustments, colour grading, posture coaching for the “after” shot, filters applied before posting to social media. These modifications are so normalised that most gym owners would not even categorise them as editing.

AI tools have pushed the line considerably further. Platforms now exist specifically to generate gym and fitness imagery, offering tools that enhance muscle definition, improve skin tone, and adjust body composition with a few clicks. Some of these platforms are marketed directly to personal trainers and gym owners as content creation solutions.

The MIT study suggests the issue extends beyond whether these images mislead people at the moment of viewing. The deeper problem is whether they alter what viewers subsequently believe they experienced. A prospect who scrolls through AI-enhanced gym transformation photos on your Instagram feed may not just feel impressed in the moment. Over time, they may come to remember bodies, including their own, differently from how those bodies actually looked.

That sits in a different category from conventional misleading advertising.

What this does not mean

The study used generic photographs, not personal images. Participants viewed someone else’s photos rather than their own transformation pictures, and the leap from “AI-edited photos of strangers create false memories” to “AI-edited progress photos change how members remember their own bodies” requires further research to confirm.

The sample was adequately powered and well-designed, but it was also online and US-based. The Instagram-style interface adds ecological validity for social media contexts without replicating the in-person dynamics of a gym consultation where a trainer shows progress photos across a desk.

The 2.05x figure is relative to a control group that also formed some false memories, because false memories occur naturally when people view images twice with a gap between viewings. The baseline matters here. The AI effect is real and statistically significant, but it is an amplifier of something that already happens rather than a phenomenon materialising from nothing.

Pre-registration and the Best Paper award speak to methodological confidence, and Loftus’s involvement means the false memory paradigm was applied by the researcher who built it. The study is strong. But the bridge from laboratory to gym floor still has gaps, and honest reporting should acknowledge them.

Research summary

The question

If the most effective gym marketing tool in existence works partly by altering what people remember rather than what they saw, what are you actually selling: the result, or the memory of a result that may never have existed in quite that form?

People also ask

Can AI-edited photos create false memories?
Yes. A 2025 pre-registered study from MIT Media Lab (Pataranutaporn et al.) found that AI-edited images significantly increased false memory formation. AI-generated videos of AI-edited images doubled the false recollection rate compared to viewing unedited images, and this occurred even when images were explicitly labelled as AI-enhanced.

Do transformation photos work in gym marketing?
Before-and-after gym transformation photos remain the most persuasive visual marketing asset for fitness businesses. Research on false memory formation suggests their effectiveness may partly stem from altering how viewers reconstruct memories of bodies and results, rather than simply demonstrating factual outcomes.

Are AI gym photos misleading?
The MIT study suggests the problem goes beyond conventional misleading advertising. AI-enhanced fitness imagery may not just misrepresent reality at the moment of viewing. It may alter what viewers subsequently believe they saw, creating confident false memories that persist after the viewing experience ends.

And if the label “AI-enhanced” does not protect the viewer, it probably does not protect the person posting it either.

References

Pataranutaporn, P., Archiwaranguprok, C., Chan, S. W., Loftus, E., & Maes, P. (2025). Synthetic human memories: AI-edited images and videos can implant false memories and distort recollection. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (pp. 1–20). ACM. https://doi.org/10.1145/3706598.3713697

Loftus, E. F., & Palmer, J. C. (1974). Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning and Verbal Behavior, 13(5), 585–589.

Fitness is BS uses behavioural science to help gym owners understand what actually drives member behaviour. Free consultation →