Finding Oracle text from a card photo
How Esper Forge combines artwork recognition, multilingual printed names and text fallback to identify physical Magic cards.
Why artwork is the strongest first signal
A phone photo rarely gives perfect text. Glare, sleeves, foils, perspective and a small title line can confuse optical character recognition. Artwork occupies a larger part of the card and remains recognizable across many translated printings, so Esper Forge compares an image fingerprint against its local Scryfall-derived artwork index before relying on words.
The comparison tolerates modest crop and framing changes. It is still a similarity search, not facial recognition for cards: alternate art, borderless treatments and heavily obscured images can point to a different printing or produce no confident match. The result is always resolved back to canonical card data before Oracle text is shown.
Text and language provide the fallback
When artwork is ambiguous, the title line becomes useful. Esper Forge's localized-name index covers the regular Magic print languages represented in the bulk data, including English, German, Italian, Spanish, French, Portuguese, Japanese, Korean, Russian, Simplified Chinese, Traditional Chinese and Phyrexian where a printed name is available.
The OCR cleanup keeps non-Latin scripts and removes common frame decoration or mana-cost noise. A localized printed name then maps back to the canonical Oracle card, so a Japanese or Russian printing can display the same current rules identity used elsewhere in a deck.
Double-faced, split and Adventure cards
Special layouts are harder because one physical frame can contain multiple names and text boxes. Transforming and modal double-faced cards have two faces; split cards share the frame; Adventure cards combine a permanent and a spell. Esper Forge treats the full canonical double-slash name as one card and can use the dominant artwork region when the secondary text box would otherwise distort recognition.
For the best result, photograph the complete front face first. If a transforming card is currently on its back face, rotate it and scan the front when possible. The large-card view will show Oracle text and the relevant faces after identification.
Take a photo the matcher can use
Fill most of the frame with one card, keep all four corners visible, avoid strong reflections and hold the phone parallel to the card. Use a neutral background rather than a playmat with similar artwork. For a binder page, scan one pocket at a time instead of asking the recognizer to choose among nine cards.
If the first attempt fails, change the angle slightly or use the manual Oracle search. No recognition method can recover information hidden by glare or an opaque sleeve, and a confident manual card name is better than repeatedly accepting a weak visual match.
Open your workspace or start with a private guest draft.
Checking…