Getting from a label image to the right track is a waterfall of decisions, each step narrowing the field, with a fallback ready whenever one comes up empty. It's also self-improving. Every scan benefits future scans for the person after you of the same item by populating metadata.
Apple's Vision framework reads every line of text on the label, building consensus across multiple camera frames, so glare, shadows or a shaky hand doesn't throw it off. It corrects for common misreads like H / N or an upside-down label.
From that raw text, Label Mate determines catalog-number candidates, artist names, and titles, splitting fine print and throwing out boilerplate. The catalog number is the strongest signal, so it leads.
Candidates go to Label Mate's catalog service in a single batched request, matched against 7.5M+ catalog numbers. It parses over all releases each one belongs to, with artist and format attached. Barcodes and album covers feed the same lookup, but more direct.
When one catalog number maps to several pressings, the artist and title read off the label break the tie. If the number turns up nothing, the cascade falls back to resolving straight from artist + title instead.
Before a song plays, the chosen release is verified against Apple Music. Fuzzy matching handles punctuation, featured artists, and various-artists compilations so you land on the real track, not a re-recording or a lookalike.
The full tracklist loads into the player. Each track streams from Apple Music; when Apple Music doesn't have it, Label Mate falls back to a YouTube video, and when you don't have an Apple Music subscription, you'll still hear a 30-second preview.
No text to read on an album jacket, so covers go through a different pipeline: cast a wide net, order the catch, then get geometric proof.
The cover is cropped out of the camera frame and distilled into a compact visual fingerprint on your phone. That fingerprint is compared against nearly two million album covers in Label Mate's index, pulling back a shortlist of look-alikes in a fraction of a second.
The shortlist gets reordered using everything else the camera saw: any words caught on the jacket are matched against artist and title, the cover's color palette is compared, and better-known releases get a nudge, so the real record rises to the top.
Before anything auto-plays, the top candidates face a stricter test: hundreds of distinctive points on your photo are geometrically matched against the real artwork. If one cover truly locks on, it plays immediately; if it's genuinely ambiguous, you pick from the top matches instead.
Deliberately AI-avoidant. There's no chatbot in this pipeline asking a generative AI "what record is this?". That costs seconds and ultimately is a guess. AI doesn't know about the local indie band. Visual fingerprints and geometric verification answer in milliseconds and prove the match. In a record store, speed is a feature.
Honest fine print: no identification system is perfect. A very worn or damaged label, an obscure private pressing, or a jacket with reissued artwork can slip past a scan. When that happens, try the cover, the barcode, or better lighting. Bright, diffused light is best. The catalog is refreshed monthly, so a record released last week may not match yet, and music that never came out on vinyl isn't in the look up at all.