Your handwritten notes→ digital flashcards

How photo-to-flashcard conversion actually works

What happens to a photo of a textbook page

A single photographed page of a textbook holds roughly 500 to 800 words. Laxu AI reads the image, extracts the text, and then reads that text for what is testable: definitions, labelled diagrams, numbered processes, and the relationships between them. It does not simply turn every sentence into a card. From one clear page you can expect 22 to 40 flashcards along with structured notes and a practice quiz, generally in under two minutes. You can photograph several pages and submit them together, which is usually better than one at a time — a process split across two pages stays intact as a single concept rather than being cut in half. JPG, PNG and HEIC all work without conversion. Screenshots of slides work exactly as well as photographs, and are usually sharper because nothing is being focused by hand.

Whether handwriting actually works

Handwritten notes work, with a caveat worth knowing before you rely on it. Neat handwriting on unlined or lightly ruled paper is read about as accurately as print. Cursive, cramped margin notes, and anything written at speed during a lecture are considerably less reliable, and mathematical notation is the hardest case of all — superscripts, subscripts and Greek letters are frequently misread. The practical approach is to photograph handwritten pages in good light, straight on rather than at an angle, and to check any card containing a formula against the original before you memorise it. Printed material, slides and textbook pages need no such checking. If your handwriting is the kind other people struggle with, assume the extraction will struggle too and check the cards once.

Photographing diagrams, tables and equations

Diagrams are handled by reading their labels and captions rather than the picture itself, so a well-labelled anatomical diagram produces good cards while an unlabelled graph produces almost none. Tables convert well when the header row is legible, because the column names become the thing each card asks about. Equations are the weakest case: a photographed equation is read as text, and a misread exponent changes the meaning entirely. If a page is mostly mathematics, you will get better results photographing the surrounding explanation and writing the equations yourself than relying on the extraction. Include the caption in the frame — it often carries the context the diagram assumes. A labelled diagram with its caption produces markedly better cards than the same diagram cropped to the picture alone.

Getting a usable photo the first time

Most poor results come from the photograph, not the processing. Shoot straight down rather than at an angle, so the text is not trapezoidal. Fill the frame with the page and leave the margins out. Avoid the shadow of your own phone falling across the text, which is the single most common problem and is fixed by moving the light source rather than the phone. Natural light beats a room light; a lamp directly overhead beats both. If the page is glossy, tilt slightly to move the reflection off the text rather than increasing brightness. A readable photo taken in five seconds beats a careful one taken in a shadow. If a page comes back with obvious gaps, retaking the photo is almost always faster than editing the cards it produced.

Best ways to turn images into flashcards, ranked

1. Photograph directly in an app that generates cards. One step, and the camera framing guidance is built in. Laxu AI does this on web, iOS and Android; free accounts get one upload, Pro is $4.99/month for unlimited.
2. Scan to PDF first, then upload the PDF. Better for long documents — a scanner app deskews the pages, and multi-page PDFs keep a chapter together as one submission instead of many separate images.
3. Run OCR separately, then paste the text. More control over correcting misreads before cards are made, at the cost of moving text between two tools every time.
4. Type the cards from the page yourself. Still the right answer for dense mathematical notation, where extraction errors cost more time to find than typing would have taken.

Frequently asked questions