Emoji Scavenger Hunt
- Brand
- Year
- 2018
- Objective
- Awareness
- Industry
- Tech & Telecom
- Mechanic
- Puzzle
- Platform
- AR & VR
- Agency
- Methodologie
Objective
Google Brand Studio built Emoji Scavenger Hunt to launch TensorFlow.js in 2018 by turning an abstract idea — machine learning running inside a phone's browser — into something anyone could feel in seconds. The brief was to keep it fast, playful, and as instant as trading emoji, while proving that heavy ML computation could run entirely on the client with no server behind it. Doing all the recognition on-device also doubled as a privacy reassurance: no camera footage was ever collected or sent anywhere.
Solution
Players open a link on their phone, grant camera access, and start playing right in the browser tab — no app to download. Each round shows an emoji of an everyday object and starts a countdown, and the player walks around a room or street pointing their camera at real things while the game calls out its guesses in text and voice. Spot the right object and it counts as a find, adding time to the clock and serving up the next emoji, with the objects getting trickier as the hunt goes — from a hand, a shoe, or a book to a banana, a candle, or a scooter. The hunt ends when the timer runs out or all ten objects are found, closing on a recap screen with snapshots of everything caught, a final score, and buttons to share the score or play again.
Results
More than 2 million emoji found by players worldwide, including 85,000 different light bulbs and 66,000 pairs of jeans identified in-game.
- 1
The player lands on a title screen explaining the premise — locate the emoji shown in the real world using the phone's camera — with a reminder to turn sound on, since the game narrates its guesses aloud; on desktop it warns the experience is built for phones.
- 2
The player taps the single "Let's Play" button, confirms an age-13+ disclaimer, and grants camera access, without which the game cannot run.
- 3
The game loads its on-device recognition model, explains the rule in one line — find the emoji and point the camera at it before time runs out — and starts a round after a short audible countdown, opening with 20 seconds on the clock.
- 4
The core screen is a full-screen live camera feed overlaid with the target emoji and a countdown timer; the player walks around pointing the camera at objects while the model continuously analyzes the feed in real time.
- 5
The game narrates and displays its current best guess as the player searches, updating every couple of seconds to signal how close the camera is to the target, and plays an urgent alert sound when the timer drops to five seconds or below.
- 6
Matching the object in frame to the target emoji triggers a success sound and a "found" message, adds ten seconds to the clock, and lets the player advance to the next emoji via a "Next Emoji" button; targets grow harder to find as the round progresses, from an easily reachable shoe or book to trickier items like a banana, candle, or scooter.
- 7
The player can quit at any time through a confirmation prompt, and the round itself ends when the timer hits zero or all ten items have been found.
- 8
A results screen shows the items found, illustrated with photos captured during the round, alongside a note that no photos are saved or sent anywhere — everything stays local to the device; a perfect or empty run gets its own congratulatory or consolation message.
- 9
From the results screen the player can share their score or start a new round, and separate About, FAQ, and Contact sections reiterate that all recognition happens locally with nothing stored on Google's servers.
