Esper Forge
/
All features
Card recognition and rulings

Move from a physical card to its current Oracle meaning

A printed Magic card is a snapshot. Oracle text, legality and rulings can change after that printing leaves the press. The Oracle workspace resolves a physical or typed clue to canonical data before presenting the answer, so a translated name or alternate frame does not become a separate rules object.

Open Oracle searchUpdated September 22, 2026

Artwork is the first visual clue

Phone photos often contain glare, sleeves, perspective distortion and title text too small for reliable optical character recognition. Artwork occupies more of the card, so Esper Forge first compares a visual fingerprint with the artwork index derived from its local Scryfall bulk data. A confident match is then resolved to the card's canonical identity.

Alternate art and heavily cropped photos remain difficult because the same card can have very different pictures. The matcher therefore treats visual similarity as evidence, not certainty. When it cannot produce a trustworthy result, text and manual search remain available instead of pretending that the closest image must be correct.

Printed language becomes a fallback, not a barrier

The localized-name index maps supported printed names back to the English Oracle identity. It can work with regular print languages represented in the bulk data, including German, French, Italian, Spanish, Portuguese, Japanese, Korean, Russian, Simplified Chinese and Traditional Chinese. OCR cleanup retains non-Latin scripts while removing common frame and mana-cost noise.

The answer follows the site's selected language where interface copy is available, but card names and Oracle text remain grounded in canonical card data. This avoids translating rules text informally and accidentally changing what a card does.

Special layouts need special handling

Transforming and modal double-faced cards have more than one face. Split cards share a frame, and Adventure cards combine a permanent with a spell. Esper Forge stores the canonical double-slash name and relevant face images together, allowing hover previews, large-card views and proxy preparation to show the correct object instead of reporting that one face is missing from the index.

For a better scan, fill most of the photo with one complete card, keep all four corners visible and avoid a background that resembles the artwork. If a double-faced card is showing its back, photograph the front when possible. Manual search is always preferable to repeatedly accepting a weak visual guess.

How a photo search is resolved

  1. 01

    Normalize the image

    The client corrects orientation and prepares a consistent region for visual comparison and OCR.

  2. 02

    Compare artwork

    A local visual index searches known printings before the title line is trusted.

  3. 03

    Read text if needed

    Localized title candidates are cleaned and matched to known printed names.

  4. 04

    Resolve canonical data

    The selected printing is connected to current Oracle text, faces, legality and available rulings.

Oracle search questions

Why can two copies of the same card scan differently?

Different artwork, foiling, language, glare and crop can change the visual evidence. A second photo or manual title search may be more reliable.

Does photo recognition replace official rulings?

No. Recognition identifies the card. The displayed canonical data and rulings provide the rules reference; unusual tournament questions may still require a judge.

Are uploaded photos used as deck preferences?

No. A recognition image is a search input and is not treated as an AI deckbuilding preference.