How BreedQuest works
Point your phone at a dog and BreedQuest gives you its best guess at the breed, right there on the screen, with a card to keep if you're right. Here's what actually happens at each step — including where a photo-based guess can be wrong, and why.
The scan
Before anything else, BreedQuest checks the photo itself, on the phone, using Apple's on-device Vision framework. It's looking for one thing: is there an animal in this photo, and is there exactly one? A photo with no dog in it, two dogs overlapping, or a shot too dark or blurry to read gets turned back with Retake or Choose another — never a guessed name from a photo that couldn't support one.
Once a photo passes that check, an on-device machine-learning model reads for the breed. It never uploads the photo to do this — the whole scan, gate and guess, happens on your phone; we cover exactly what that means for your photos on what happens to your photos. The model returns a read on its own confidence, not just a name: sure when it's confident in one breed, mix when it thinks it's looking at a blend, and unsure when the photo doesn't give it enough to commit to either.
Scanning works on dogs today — 118 of the 458-breed catalog, the breeds the model was actually trained to recognize. Cats aren't scannable: their 72 profiles live in the catalog to browse, the same as every breed before it has a scan behind it.
The card
A breed you scan becomes a card: a gold-foil illustration with a serial number and a rarity tier, from Common up through Legendary. Purebreds draw from a shared library of illustrated components, so two Labradors' cards look alike — the way two real Labradors do.
Mixed breeds are the exception, and the app is built to make that visible. When BreedQuest reads a dog as a blend, it generates a card that mixes both parent breeds' visual traits, so no two mixed-breed cards are the same — see the mixed-breed hub. That's deliberate: appearance is the weakest signal for a mix's real ancestry (more on that below), and a one-of-a-kind card is more honest about what a mixed-breed guess actually is than pretending it landed on one clean breed.
The collection
Every card you unlock joins your collection, and every breed — scanned or not — has a full page behind it: temperament, history, care notes, the works. None of that needs an account. Scanning, browsing the 458-breed library and reading breed guides are free with nothing to sign up for.
An account only enters the picture if you want to keep records for your own dog or cat — routines, health notes, a passport-style profile, backed up and available on the web. That's free too, for one pet; see everything the app does, plan by plan, on the features page.
How accurate is it, honestly
A breed scan is a probability, not a certainty. It's a machine-learning model reading a photo, and it can be wrong — most often on mixed-breed dogs, where there's no single clean answer to draw a line around in the first place. That's not a limit specific to us: a 2022 study in Science on shelter dogs found that people, looking at a photo, identify only a fraction of a mixed dog's actual genetic ancestry by sight. Appearance is a genuinely weak signal for ancestry, for a person or a model, and that's the honest ceiling on what any photo-based guess can tell you.
The photo you take changes the result more than people expect. The app teaches three rules before your first scan: one dog, whole dog — clear and well-lit, the entire body in frame; head to the camera, tail visible — profiles and sleepy close-ups confuse the scan; and nothing else in the shot — toys, laps and leads steal the spotlight. A dog mid-play, in costume, or photographed face-only with its eyes shut trips the model up in ways a plain, full-body photo won't.
So don't treat a scan as the last word on anything that matters — insurance, adoption paperwork, a breed-specific rule where you live. It's a fun, honest best guess that says "sure", "mix" or "unsure" rather than pretending to know more than a photo can show.
FAQ
How does BreedQuest identify a dog's breed?
An on-device machine-learning model looks at your photo and matches what it sees against the breeds it was trained on, then returns its best guess with a confidence read: sure, mix, or unsure. The photo is checked and processed entirely on your phone; nothing is uploaded to identify a breed.
How accurate is the breed scanner?
It's a probability, not a certainty, and it's most confident on a clear, full-body photo of one dog. Mixed-breed dogs are the hardest case for any appearance-based guess, ours included — research on shelter dogs has found that people identify only a fraction of a mixed dog's true ancestry by sight, and a photo doesn't tell a model much more than it tells a person.
Does BreedQuest work on cats?
Not for scanning, no. The camera-based breed scanner works on dogs — a trained subset of the catalog today. Cat breeds are in the catalog to browse and read about, the same as every dog breed, but there's no cat scan yet.
What does "unsure" mean when I scan a dog?
It means the model didn't have enough in the photo to commit to a breed or a blend with any confidence — usually the angle, the light, or something else in the frame. Retake the photo following the app's three rules (whole dog, head to camera, nothing else in shot) rather than treating a low-confidence guess as a real answer.
Do I need an account to scan?
No. Scanning, the breed card and the whole breed library work with no account and no sign-up. An account only comes in if you want to save your own pet's records with cloud backup, and that's free too, for one pet.
When can I download BreedQuest?
BreedQuest is still in development, with a 2027 launch. Join the waitlist below and we'll write once, the day it's live on the App Store — nothing else, ever.
More on what happens to the photo itself: what happens to your photos →