The researchers at MIT asked a question that sounds trivial but turned out to be deeply counterintuitive: can how memorable an image is be measured? The answer was yes. Almost everything we thought mattered fell away. Beauty predicted nothing. Color intensity predicted nothing. Yet thousands of strangers around the world remember exactly the same images and forget exactly the same images. If you’re choosing a poster for your wall, this means something unexpected: the subject you’ll remember isn’t necessarily the one you’ll love living with.
Over the past fifteen years, a field of research has emerged around what is known in English as image memorability. It’s not about your memory, but the image’s: a quality of the subject itself that determines how many of us recognize it when we see it a second time. The field has produced findings that are hard to reconcile with how we usually talk about art, and are therefore worth looking at more closely before you order your next print.
MIT’s 60,000 images: how memory became measurable
It began with a study from MIT’s Computer Science and Artificial Intelligence Laboratory, CSAIL, published in 2011 by Phillip Isola, Antonio Torralba, and Aude Oliva. Their method was cleverly simple. Participants watched a long stream of images and pressed a key whenever an image was repeated. No one knew which images were test images. No one was asked to memorize anything. Yet it was possible to calculate exactly what proportion of participants recognized each individual image.
The result was that the images fell into a stable ranking. Some subjects were recognized by almost everyone. Others passed before the eye and left nothing behind. When the researchers split the participants into two random halves and compared the halves’ rankings, they found a correlation of around 0.75 for mixed subjects. In other words, two groups of strangers who had never met agreed on which images were worth remembering.
In 2015, Aditya Khosla, Akhil Raju, Torralba, and Oliva scaled up their work to create LaMem—the largest annotated database of image memorability at the time, with 60,000 images from widely varied sources. Using a neural network they called MemNet, they achieved a rank correlation of 0.64 with people’s actual memory performance, compared with an inter-rater agreement of 0.68 among people. The machine came almost as close as we do to one another. The study was published at the ICCV 2015 conference, and the database is still freely available.
Quick facts: The MIT CSAIL LaMem database contains 60,000 images with measured memorability. The MemNet model, presented at the ICCV 2015 conference (DOI 10.1109/ICCV.2015.275), predicts memorability with a rank correlation of 0.64 - compared with 0.68 between human viewers. Among face photographs, agreement between different groups is 0.68.
Half of memory is in the image - not the viewer
The field's most transformative conclusion is quantitative. When researchers break down the variation in what people remember, it turns out that about half is explained by the image itself. The other half is distributed across the viewer, the situation, fatigue, distractions in the room, and all the other noise that makes up life.
That is a remarkable distribution. We are used to thinking of memory as something personal, shaped by our own experiences and our own history. Image memorability research says that this is only half the story. The other half is a property of the image, as measurable as its width in centimeters.
Wilma Bainbridge, now a researcher at the University of Chicago, showed back in 2013 in the Journal of Experimental Psychology: General that the same is true of faces. We remember the same faces as one another, regardless of whether we find them attractive. In a follow-up series of five psychophysical experiments, published in Neuropsychologia in 2020, Bainbridge and Jesse Rissman tested whether the effect was really just attention in disguise. They tested bottom-up attention, top-down attention, and so-called priming. None of these could explain away memorability. Memorable images do not automatically draw the eye. They stay with us anyway.
"About half of what you will remember from an image was determined before you ever saw it."
What does not predict memorability: beauty, color, and brightness
This is where things get uncomfortable for anyone selling art. Research has repeatedly failed to link memorability to the qualities we normally use to judge an image. Aesthetic quality does not predict it. Emotional charge does not predict it. Brightness and color saturation do not predict it. Nor do the basic image features that computer vision has long relied on, such as spatial frequencies or color histograms.
Bainbridge summarized the field in a review article in Psychology of Learning and Motivation in 2019, titled "Memorability: How what we see influences what we remember". The conclusion was that memorability is stubborn. It cannot be reduced to any single attribute we typically measure.
That means a poster can be exquisitely beautiful and surprisingly easy to forget at the same time. And vice versa: an image no one would call beautiful can burn itself into your memory. If you’ve ever wondered why an old advertising poster from your childhood stays in your mind while a much more skillfully painted landscape you saw last month is already gone, this is the explanation. Skill and memorability are two separate axes.
Faces, animals, and the unusual: the semantics behind images we don’t forget
If it’s not color or beauty, then what is it? The answer researchers have been moving toward is semantic rather than visual. It’s about what the image depicts and how quickly the brain can determine that.
A study published in the journal Imaging Neuroscience in 2024 linked memorability to both faster high-level perceptual processing and more efficient semantic activation. Memorable images are simply easier to perceive and easier to put into words. Analyses of MemNet’s internal layers have identified which object categories boost memorability and which reduce it. Faces and animals rank highly. Open expanses and large, uniform areas rank low.
A 2025 working paper added one piece to the puzzle: images that are difficult for a model to reconstruct are more memorable, and uniformity across an image is negatively associated with memorability. In other words, distinctiveness seems to be a driving force. An image that stands out from a crowd of similar images is stored more deeply. A study in Scientific Reports in 2025 pointed to the same mechanism from the other direction, showing how interference between similar impressions wears down memory.
That explains a familiar frustration. A series of similar botanical prints can look exquisitely beautiful on the wall and yet blur into a single fuzzy whole in your memory, while one contrasting image in the same room remains razor-sharp. If you’d like to read more about why some images feel beautiful without necessarily sticking in your memory, we’ve covered that research in our article on processing fluency and the experience of beauty.
Tips: Test the distinctiveness principle at home before you buy. Place your chosen poster among five other prints you already own, look at it for ten seconds, leave the room, and describe it to yourself an hour later. If you can do so effortlessly, the image has the distinctiveness measured by the research. If you need to go back and check, that’s a sign the image is competing with the rest of the wall instead of complementing it.
Memorability isn’t the same as liking something – and that determines how you should choose
Now comes the practical twist, and it runs counter to how research findings are usually applied in interior design. The fact that a motif is memorable is not automatically a reason to hang it in the bedroom.
Where memorability actually comes in handy
In the hallway, stairwell, and entrance, you see the wall for a few seconds at a time, often while moving. There, a memorable motif works in your favor. It gives the room an identity that guests carry with them. The same applies to the wall behind your desk during a video call and the only wall in a small room. A distinctive motif with a clear figure, preferably a face or an animal, makes more of a few seconds of viewing than a subtle color study.
Where memorability can become a burden
In rooms where you spend a lot of time, almost the opposite applies. A motif that shouts every time you pass by can become tiring over the years. There, repeated discovery matters more than immediate impact, and abstract and atmospheric motifs often work better. We’ve written about this mechanism in our overview of the mere exposure effect and the art you learn to love. It’s also worth combining this with insights into how long we actually look at art, since viewing time determines which of the mechanisms gets to work.
How to build a wall where both can coexist
The most reasonable interpretation of the research is to mix intentionally. Let a distinctive anchor motif carry the memorability. A striking face or a defined silhouette from our collection of flower faces does this well. Then let the rest of the wall be more subdued: soft abstract motifs or calming nature motifs that you can live with without them demanding anything. That way, you get both impact and staying power, instead of sacrificing one for the other.
Frequently asked questions about image memorability and choosing a poster
Image memorability is one of the few areas where research can actually say something measurable about images on the wall. But what it measures is recall, not love. The wisest approach is to let a single image carry the memory and let the rest be quieter. At Bozetto, we print everything in Sweden and offer a 60-day return policy, so you can try out how your anchor image works in your own light. Feel free to start with our bestsellers – the images that have already proved most popular.