AI transcription has made digitizing a paper scoresheet almost effortless: snap a photo, and you get SAN moves and a downloadable PGN in seconds. But one rule decides how good the result is, and it has nothing to do with the AI. It's garbage in, garbage out. A crisp, well-lit photo transcribes almost perfectly; a blurry, shadowed, crumpled one forces you to hand-correct half the game.

The good news: a clean photo takes about thirty seconds once you know what matters. This guide covers the factors that affect OCR accuracy, from room light to how you hold your phone, plus how to review the recognized moves before you export.

Why Photo Quality Decides Everything

Optical character recognition reads pixels. When those pixels are sharp and high-contrast, the model sees clean strokes and reads them confidently. When they're soft, dim, or distorted, it has to guess, and that's where errors creep in: an Nf3 becomes Nf5, a castle gets dropped, a whole row blurs into noise. You can't change your handwriting after the game, but you can control the photo. That's the lever with the biggest payoff.

Lighting: Even and Bright Beats Everything

Light is the single most important variable. Aim for bright, even, diffuse light across the entire sheet.

• Use soft, ambient light. Daylight near a window (not direct sun) is ideal; overhead room lighting works too.
• Kill harsh shadows. Don't let your hand, phone, or head cast a shadow across the page. Position yourself so the light isn't blocked by you.
• Avoid glare. A glossy sheet or laminated clipboard under a single bulb creates a hotspot that wipes out the moves underneath. Tilt the sheet or move the light until the glare is gone.
• Skip the flash unless the room is dark. On-camera flash usually creates a harsh central glare spot.

If half the page is in shadow and half in bright light, OCR struggles across that seam. Even illumination matters more than raw brightness.

Flatten the Sheet

Folds, curls, and creases distort the lines of text and throw off recognition.

• Smooth the sheet flat on a table before shooting.
• If it was folded in a pocket, press it under a heavy book for a minute first.
• Hold curled corners down, or weight them, so the whole surface sits in one plane.

Frame It Straight-On and Fill the Frame

How you aim the camera matters as much as the light.

• Shoot straight down, directly above the sheet, camera parallel to the page. Angled shots stretch the rows and make columns lean, which confuses move-pair alignment.
• Capture the whole sheet, header to result. Don't crop off the top or the bottom rows.
• Fill the frame. Get close enough that the scoresheet occupies most of the photo. More pixels per move means better reads; a tiny sheet in a big photo wastes resolution.
• Keep it level so the rows run straight across.

Focus and Sharpness

A blurry photo is the most common cause of bad transcription.

• Tap to focus on the text before you shoot, and wait for it to lock.
• Hold the phone steady or brace your elbows on the table. Motion blur is invisible on a small preview but ruins OCR.
• Check the photo at full zoom before uploading. If you can't read a move clearly, the AI can't either. Reshoot.

Help the Ink Help You

A few habits pay off later:

• Dark ink on the original (black or dark blue) reads far better than light pencil or faded pen.
• Plain, contrasting background. A dark table for a white sheet, for example. Avoid busy patterns, other papers, or clutter in the frame.
• One sheet at a time. If your game ran across multiple pages, photograph each page as its own separate, clear shot rather than cramming two into one frame. Upload them in order.

A Quick Pre-Upload Checklist

Run through this before you hit upload:

• Bright, even light, no harsh shadow across the page
• No glare hotspots on the sheet
• Sheet flattened, no folds or curl
• Camera straight down, parallel to the page
• Whole sheet in frame, header to result
• Scoresheet fills most of the photo
• Tapped to focus, image is sharp at full zoom
• Plain, contrasting background
• Multi-page games shot as separate photos

Review the Moves Before You Export

Even a great photo deserves a quick check, and ScanChess makes this easy. After OCR you don't just get raw text, you get move-legality validation. The system checks each move against the rules of chess and flags anything illegal or ambiguous, so a misread move that can't legally be played gets highlighted instead of silently corrupting your PGN.

Here's the review flow:

1. Look at the flagged moves first. A flag usually means the AI misread a single character, easy to spot and fix once you compare it to your paper sheet.
2. Edit any move in place, then re-check that the game still plays through cleanly.
3. Step through the interactive replay. If it plays naturally, your transcription is solid.
4. Export the clean PGN once everything checks out.

Legality flagging is your safety net, but it works best when the photo gave the model good data to start with. If you're new to the page layout, our guide on how to read a chess scoresheet helps you match recognized moves to the right rows, and the rundown of common chess notation mistakes will help you spot recurring errors faster.

Put It to Work

Clean photos plus legality checks turn a stack of paper into accurate, shareable PGN files with almost no manual typing. The whole photo to PGN pipeline only works as well as the image you feed it, so spend the extra thirty seconds to get the shot right.

Ready to digitize your games? Upload a scoresheet photo and let the scanner do the work at ScanChess board recognition, where you can also turn a board photo into FEN.