Google Arts & Culture "Say What You See": A Guide
Google's "Say What You See" turns describing a painting into a guessing game between you and a machine — and it's far more revealing than it sounds.
Most of us look at art for about eight seconds before drifting on. "Say What You See," one of the more disarming experiments inside Google Arts & Culture, exists to break that habit. It hands you a piece of art, hides the title, and asks you to describe it in plain words until the machine guesses what you're looking at. It sounds trivial. In practice, it forces you to actually look — to notice the dog in the corner, the cracked window, the strange green in the sky — in a way no museum wall label ever manages.
This guide walks through exactly how the game works, how to play it well, the kinds of images you'll meet, and how it fits into the larger sprawl of free tools Google has quietly built on top of the world's museum collections.
What "Say What You See" actually is
"Say What You See" is a browser-based game built by the Google Arts & Culture Lab. You're shown an artwork — a painting, a photograph, a sculpture, sometimes an artifact — with its identity concealed. Your job is to type short descriptions of what's in the image. After each guess, an AI model responds by surfacing the artwork (or a similar one) that best matches your words. The closer your description steers the machine to the actual piece on screen, the better you've done.
The twist is that you're not naming the painting. You're describing it. "A woman in a blue dress holding a baby" is a description. "Madonna and Child" is a title — and titles are precisely what the game withholds. This reframing is the whole point. You stop trying to be clever and start being observant.
How does "Say What You See" work?
Under the hood, the game runs on a text-to-image matching model. Google has digitized millions of artworks from partner institutions, and each image has been processed so that its visual content can be compared against natural-language descriptions. When you type a phrase, the model converts your words into the same mathematical "embedding space" the images live in, then finds the artwork whose visual features sit closest to your description.
That's why specificity wins. Type "tree" and the model has tens of thousands of trees to choose from. Type "a single bare tree on a snowy hill under a grey sky" and you've narrowed the field dramatically. The game is, in a quiet way, training you to describe like a curator: composition, color, mood, and the odd telling detail.
Playing it well: a few practical tips
- Lead with the subject, then the setting. "A man on horseback" before "in a battlefield at dusk." The model anchors on nouns first.
- Name colors precisely. "Ochre" and "crimson" carry more signal than "yellowish" and "red."
- Describe what's unusual. If there's a clock melting, a person with two faces, or a fish where a head should be, say so. Oddities are great discriminators.
- Add medium and era cues if you can. "An oil painting" versus "a black-and-white photograph" instantly slices the dataset.
- Don't over-interpret. The model responds to visible content, not symbolism. "Grief" matches less reliably than "a kneeling figure with hands over their face."
Say what you see: examples worth trying
The fastest way to understand the game is to imagine a few rounds. Say the hidden image is Van Gogh's The Starry Night. A weak guess: "a night scene." A strong guess: "a swirling blue night sky with a bright yellow moon over a small village with a tall dark cypress tree." The second description hits color, motion, the cypress, and the town — exactly the features that make the painting unmistakable.
Try another. A hidden Vermeer interior. Weak: "a woman in a room." Strong: "a woman in yellow near a window, soft daylight, reading or pouring something." You've given the model light direction, palette, and an action — far more than a generic interior could be.
These rounds are weirdly addictive because they expose your own blind spots. You'll discover you barely registered the background, or that you can't name the difference between teal and turquoise, or that you describe people far more readily than objects. That gap between seeing and saying is the real lesson, and it's one reason I keep recommending the game to people who insist they "don't get" art. If you've ever felt locked out of galleries, it's worth remembering how much creative noticing you already do every day — a point I've argued at length in You're More of an Artist Than You Think.
The bigger family of Google art experiments
"Say What You See" doesn't stand alone. It belongs to a long line of playful machine-learning toys the Arts & Culture Lab has released to make collections feel less like a database and more like a sandbox. There's Art Selfie, which matches your face to historical portraits. Blob Opera, which lets you conduct four singing blobs trained on real opera voices. Art Coloring Book, Puzzle Party, Assistant on Display, and a steady drip of others.
What unites them is a refusal to treat digitized art as homework. The arts-and-technology magazine Colossal has long tracked how these tools blur the line between archive and playground, and that's exactly the spirit here: take a serious cultural resource and find the side door that invites people in through curiosity rather than obligation.
How to use Google Arts & Culture
If "Say What You See" is your entry point, it's worth knowing the rest of the platform, because the game is one feature among hundreds.
On the web
Go to artsandculture.google.com. The homepage is a feed of stories, collections, and experiments. Use the menu to browse by:
- Artists — alphabetical, with timelines and linked works.
- Mediums — painting, sculpture, photography, textiles, and more.
- Movements — Impressionism, Cubism, Surrealism, and dozens beyond. If the labels feel abstract, it's worth understanding why these categories still shape how we see, something I dug into in Why Art Movements Still Matter.
- Museums — over 2,000 partner institutions, many with Street View walkthroughs of their actual galleries.
The "Play" or "Experiments" section is where you'll find "Say What You See" alongside the other interactive tools.
On the app
The free Google Arts & Culture app (iOS and Android) packages the same content plus the camera-based features that work best on a phone. Art Selfie, Art Filter (augmented-reality try-ons of artifacts and masks), and Pocket Gallery all live here. The app is genuinely the better home for the more experimental, camera-driven toys.
Search and high-resolution zoom
One underused feature: many works are scanned at gigapixel resolution through Google's Art Camera. Open a Vermeer or a Bruegel, hit the zoom, and you can inspect individual brushstrokes and craquelure you'd never get close enough to see in person. This is where the platform stops being a toy and becomes a serious study tool.
Can I take a photo of something and have Google tell me what it is?
Yes — but that's a different tool from "Say What You See." For identifying real-world objects, artworks, plants, landmarks, or products, you want Google Lens, built into the Google app, Google Photos, and Chrome. Point your camera at a painting in a museum (where photography is allowed) and Lens will often surface the title, artist, and related results.
The distinction matters: Lens does the describing for you, while "Say What You See" makes you do the describing and grades how well you communicated. One is a utility; the other is a game about your own perception. They use related technology — visual matching — pointed in opposite directions.
How do I find my look-alike on Google Arts & Culture?
This is the feature that originally went viral, and it's still the most popular reason people download the app. It's called Art Selfie, and here's how to use it:
- Open the Google Arts & Culture app.
- Tap the camera icon at the bottom of the screen (or scroll to the Art Selfie card).
- Allow camera access and take a selfie when prompted.
- Wait a few seconds while it compares your face against tens of thousands of portraits in partner collections.
- Swipe through your matches, each with a similarity percentage and a link to the original artwork.
The matches range from flattering to comic, but the genuinely useful part is the trail it leaves: each look-alike is a real, attributed work you can then explore, zoom, and read about. People arrive for the novelty selfie and leave having discovered a seventeenth-century portraitist they'd never heard of. Note that Art Selfie is currently app-only and not available in every country, owing to regional rules on biometric data.
Why this small game is worth your time
It would be easy to file "Say What You See" under "cute AI demo" and move on. I'd push back. Describing art out loud is a skill, and it's the same skill that makes you a better looker, a better buyer, and a better talker about culture in general. The moment you have to articulate why a painting works — the diagonal that pulls your eye, the single warm note in a cold composition — you've stopped consuming art and started reading it.
That fluency pays off everywhere. It makes museum visits richer. It makes the strange backstories of canonical works — the kind collected in The Strange Stories Behind Five Famous Paintings — land harder, because you can actually see what you're reading about. And if the game nudges you toward wanting art of your own, the appetite it builds is cheaper to feed than you'd think; there's a whole world of collecting on a real budget waiting once you start trusting your own eye.
So play a few rounds. Lose a few. Notice how often the machine guesses correctly the moment you mention the one detail you almost left out. That detail — the thing you almost didn't say — is usually where the whole painting lives.