Fix low-quality OCR before approval

Outcome you are aiming for

Some pages are not worth auto-processing twice—know when a new photo beats another automated pass.

Preconditions

• Source photos or screenshots are readable enough to justify OCR.
• You can spend time on a review pass—this workflow is not one-click magic.
• You know which language context applies (Thai layouts).

Step-by-step workflow

1. Upload the pages you already trust as study material.
2. Let extraction propose draft cards from the capture.
3. Edit fronts, backs, readings, and tags until each card is unambiguous.
4. Reject weak cards instead of letting them pollute the deck.
5. Approve a tight batch, then continue with the next capture when ready.
6. Export approved cards when you want a portable Anki package.

QA checklist

• Every card has a clear question and a single best answer.
• Readings and tone marks match your learning standard.
• Duplicates and near-duplicates are merged or removed.

Troubleshooting

• Blurry photos: recapture with steadier lighting before re-running extraction.
• Mixed scripts: split cards so each tests one decision.
• Too many new cards: lower weekly intake so reviews stay honest.

Related reading

• Pillar: Import textbook photos into draft flashcards
• Next in lane: Continue this lane: Build a weekly vocab deck from class notes
• Cross-template A: Compare approaches: One-click generators vs editable review systems
• Cross-template B: Key concept: Cloze vs basic card formats