Skip to main content

Guide

Handwritten Notes to Text: Phone Photo OCR That Keeps the Original

A phone photo and an OCR step get you 90 percent of the way. The last 10 percent is proofreading numbers and names, rebuilding the structure the page had, and keeping the original image next to the text. This page covers the tools and the workflow, with examples of what the result should look like.

Updated 2026-10-27

How to convert

  1. Photograph the page properly

    Flatten the page, shoot straight down, use even light and fill the frame with one page. On a whiteboard, stand square to the board and take one photo per section if the board is wide. Check the photo at full zoom before you leave the room: if you cannot read it, the software cannot either.

  2. Pick an OCR route

    Built-in: on iPhone, Live Text lets you select text inside a photo; on Android, Google Lens does the same. Google Keep's Grab image text pulls text from a photo into a note. Google Docs opens an image from Drive with the recognized text under it. Dedicated: Pen to Print and Transkribus specialize in handwriting, including cursive and historical scripts. Stylus apps: Goodnotes (lasso, Convert to text), OneNote (Ink to Text) and Nebo convert handwriting written on the device, not photos. In Scribelet: attach the photo to a note and the text is extracted for you.

  3. Proofread against the photo, numbers first

    Handwriting OCR fails on exactly the characters that matter: 3 and 8, 1 and 7, 0 and 6, S and 5, units and decimal points. Read every number in the output against the photo. Then names, then dates. Prose errors are usually obvious; a wrong measurement is not.

  4. Rebuild the structure

    A page has headings, indentation, circled items, arrows and margin notes. Flat OCR output has none of that. Add Markdown headings for each section, turn lists back into lists, make checkboxes - [ ], and put margin comments under their own heading so you can tell your reaction from the lecturer's words.

  5. Keep the original attached

    The text is a convenience; the photo is the record. Keep them together so a doubtful word can be checked in seconds. In Scribelet the photo is an attachment on the same note; in a folder of files, link the image from the Markdown.

  6. Tag and link it so you find it again

    Add the course, project or client tag and a link to the related note ([[Harlow Creek]], [[BIO 204]]). A transcribed page with no tag is only slightly more findable than the paper it came from.

5 examples

What this looks like for different setups

The conversion is the same; what you do with the result depends on where the notes came from.

1.Whiteboard photo from a sprint planning session

When it fits: Use this when the whiteboard is about to be erased, four people wrote on it, and the arrows and columns carry as much meaning as the words.

Whiteboard photo from a sprint planning sessionExample 1

Sprint 41 planning board, 2026-10-26

Source: whiteboard photo, attached. Room 3B, 14:10. Present: Ana, Rui, Petra, Sam.

Column: Committed

  • Payment retry queue (Rui) [5]
  • Invoice PDF rendering fix (Petra) [3]
  • Onboarding email copy (Sam) [2]

Column: Stretch

  • Admin audit log export (Ana) [5]
  • Rate limit dashboard (Rui) [3]

Arrows and margin notes

  • Arrow from "Payment retry queue" to "Rate limit dashboard": Rui wrote "needs same metrics". Do the queue first.
  • Circled at the bottom: "Demo Thursday 11:00". Confirmed by Petra.
  • Unreadable word after "audit log": looks like "CSV?" [?]

Capacity

18 points committed, 8 stretch. Petra out Friday.

  • Confirm the CSV question with Ana @due(2026-10-27)
  • Book the demo room @due(2026-10-28)

2.Two pages of lecture notes with margin comments

When it fits: Use this when you take notes by hand in class and want them typed before the exam. The margin comments are the valuable part; keep them separate from the lecture body.

Two pages of lecture notes with margin commentsExample 2

BIO 204, lecture 14: Action potentials, 2026-10-22

Source: two notebook pages, photos attached. Prof. Okafor.

Lecture body

  • Resting potential about -70 mV. Maintained by Na+/K+ pump (3 Na+ out, 2 K+ in).
  • Threshold about -55 mV. Below it, no spike. Above it, all-or-nothing.
  • Depolarization: voltage-gated Na+ channels open. Peak near +30 mV.
  • Repolarization: K+ channels open, Na+ channels inactivate.
  • Hyperpolarization undershoot, then back to rest. Absolute refractory period during inactivation.

Margin comments (mine)

  • "Why 3 out 2 in?" Net negative charge inside. Ask in section.
  • "Exam Q last year was on refractory period." Confirmed with Dana.
  • Diagram of the spike was drawn, not transcribed. See photo 2.

Proofreading notes

  • OCR read "-55 mV" as "-SS mV". Fixed against the photo.
  • "Okafor" came out as "Okafer". Fixed.

#bio204 #lecture #exam-prep

3.A handwritten recipe card in cursive

When it fits: Use this when the handwriting is old, cursive and personal. Accuracy will be lower, so the photo stays as the record and the text exists so you can search and scale it.

A handwritten recipe card in cursiveExample 3

Aunt Marisa's lemon olive oil cake

Source: index card, both sides, photos attached. Written around 1988. Her handwriting, cursive, pencil on the back.

Ingredients (as written)

  • 3 eggs
  • 1 cup sugar
  • 3/4 cup olive oil (she underlined "good")
  • 1/2 cup milk
  • zest of 2 lemons, juice of 1
  • 1 1/2 cups flour
  • 2 tsp baking powder
  • pinch salt

Method (as written, lightly punctuated)

Beat eggs and sugar until pale. Add oil slowly, then milk, zest and juice. Fold in flour, powder and salt. Loaf tin, buttered and floured. 350 F about 45 minutes, "until the knife comes out clean and the top cracks".

Back of the card

"For Lena's birthday use 4 lemons" [pencil, faint]. One line is unreadable [?].

Transcription notes

  • "3/4" was read as "34". Fixed.
  • The oven temperature was read as "850 F". Fixed to 350 F.

#recipes #family

4.A field notebook from a site visit

When it fits: Use this when you wrote outdoors on a clipboard, with measurements, sketches and abbreviations only you understand. Expand the abbreviations while you still remember them.

A field notebook from a site visitExample 4

Site visit: Harlow Creek culvert, 2026-10-19

Source: 3 clipboard pages, photos attached. Weather: light rain, 11 C. On site 09:40 to 12:15 with M. Ruiz (county).

Measurements

PointReadingNote
Upstream invert142.31 mre-shot twice, matched
Downstream invert142.08 m
Barrel diameter1,200 mmcorrugated steel
Sediment depth, inlet310 mm"sed 31" in my notes

Observations

  • Scour pool downstream, roughly 2 m by 3 m, deeper than the April visit.
  • Headwall crack, hairline, left side looking downstream. Photo 7.
  • Beaver activity 40 m upstream. Ruiz wants it in the report.

Abbreviations expanded

  • "DS" downstream, "US" upstream, "HW" headwall, "sed" sediment depth.

Still to check

  • Compare inverts with the 2024 survey @due(2026-10-24)
  • Send Ruiz the beaver note for the county file @due(2026-10-21)

#fieldwork #harlow-creek

5.A wall of sticky notes after a workshop

When it fits: Use this when 40 sticky notes are grouped on a wall and the groups are the point. One photo per cluster, then one heading per cluster.

A wall of sticky notes after a workshopExample 5

Customer support retro, sticky-note wall, 2026-10-23

Source: 6 photos attached, one per cluster. 38 notes total. Facilitator: Jonah.

Cluster: Slow first response (11 notes)

  • "Queue not sorted by wait time"
  • "Weekend coverage is one person"
  • "Macros out of date, so we write from scratch"
  • Eight more saying the same thing in different words.

Cluster: Billing confusion (9 notes)

  • "Customers do not know if they are on monthly or annual"
  • "Refund policy not linked in the receipt"
  • "Proration math nobody can explain"

Cluster: Things that work (7 notes)

  • "Handover notes between shifts"
  • "Friday bug triage with engineering"

Cluster: Unsorted (11 notes)

  • Mixed. Jonah to re-sort by Tuesday.

Votes (dots on notes)

  • Most dots: "Queue not sorted by wait time" (9). Second: "Macros out of date" (6).

  • Jonah re-sorts the unsorted cluster @due(2026-10-28)

  • Proposal for queue sorting to Priya @due(2026-11-03)

#retro #support

Three ways handwriting becomes text

There are only three routes, and which one you need depends on where the handwriting is.

Written on a device. If you wrote with a stylus in Goodnotes, OneNote, Nebo or Samsung Notes, the app already has the stroke data. Select the handwriting and convert it in place. The layout survives, accuracy is high and this page is mostly not for you.

Written on paper or a whiteboard. You have a photo, so you need OCR. The built-in options (Live Text on iPhone, Google Lens on Android, Grab image text in Google Keep, open-with-Docs in Google Drive) are free and fine for print. Dedicated handwriting tools (Pen to Print, Transkribus) cost money or time and do better on cursive. Scribelet runs OCR on any photo you attach to a note and keeps the photo with the text.

Written long ago, by someone else. Old letters, recipe cards, a parent's journals. This is the hardest case: cursive, faded ink, unfamiliar abbreviations. Use a handwriting-trained tool, expect to proofread every line, and treat the text as a finding aid for the photographs rather than a replacement.

What survives and what does not

OCR is good at words and bad at everything else on a page.

  • Words in print. Survive well in good light. Expect a handful of errors per page.
  • Numbers. The weak point. 3 and 8, 1 and 7, decimal points and units are confused constantly. Check every one.
  • Structure. Headings, indentation and columns are flattened. A two-column board comes out as interleaved lines. You rebuild this by hand.
  • Arrows, circles, underlines. Gone. These usually encode the decision or the emphasis, so write them out as sentences while you remember.
  • Diagrams and sketches. Not text. Keep the photo, write a caption.
  • Margin notes. Captured as words but merged into the body. Separate them under their own heading or you will later quote your own reaction as the lecturer's claim.

The practical consequence: the conversion is a five-minute job for the software and a ten-minute job for you. Budget the ten minutes. A transcription nobody proofread is a liability the first time someone acts on a wrong number.

How to proofread in ten minutes

  1. Read the numbers against the photo. All of them, including dates.
  2. Read the names. OCR produces plausible wrong names, which are worse than obvious garbage.
  3. Scan for merged or split words and for lines that belong to a different column.
  4. Mark anything you still cannot read with [?] rather than guessing.
  5. Write a two-line transcription note listing what you fixed. Your error pattern repeats; next time you know where to look.

Whiteboard photos specifically

A whiteboard is the case where speed matters most, because the board gets erased. Do three things before you leave the room: photograph each section square-on, write down who was there and the date, and write one line per arrow describing what it connected. The words on the board are rarely the decision. The arrow between "payment queue" and "rate limit dashboard" was the decision, and no OCR engine will tell you what it meant.

Then put the photo and the text in the same note, tag it with the project, and turn the action items into dated tasks. The first example above is the full shape.

When not to convert

Not every page earns transcription. A scratchpad for a problem you solved, a doodled agenda, a shopping list you already bought: photograph for the record if you want, and leave it as an image. Convert pages that hold decisions, measurements, study material or something you will quote back to another person. If you are also dealing with recorded audio rather than paper, the lecture recording to notes page covers that route.

Cleanup tips

  • One page per photo, straight on

    OCR engines handle one flat page well and two curved pages badly. Shoot each page separately, square to the page, in even light without a shadow from your hand or phone. Dark ink on light paper gives the best contrast; for faded pencil, bump the contrast before uploading.

  • Mark what you could not read

    Put [?] after any word you guessed. It is faster than agonizing, it is searchable later, and the photo is attached if you ever need to try again. A transcription with honest gaps beats a confident wrong one.

  • Keep a transcription-notes section

    Write down the errors you fixed: a 3 read as an 8, a name misspelled. It takes 30 seconds and it tells you which kinds of errors your handwriting produces, so you check those first next time.

  • Whiteboards: capture the arrows in words

    OCR reads words, not arrows. Before you erase, write one line per arrow: what points to what and what it meant. The words on the board are rarely the decision; the arrow between two boxes usually is.

  • Do not transcribe diagrams

    A sketch, a graph or a circuit stays as a photo. Write a one-line caption so it is searchable, link to the photo, and move on. Describing a diagram in prose costs ten minutes and produces something worse than the picture.

  • Skip pages you will not reread

    Not every page deserves OCR. If a page is a scratchpad for a problem you solved, photograph it for the record and leave it as an image. Convert the pages that hold decisions, measurements, or material you will study.

In the app

How to do this in Scribelet

  1. Open the note, open the Attachments section (the paperclip) and add the photo, or tap the camera button when composing to start a new note from an image. Photos and PDFs are both accepted.

  2. A vision model extracts the text (OCR) and writes a caption. Both are indexed, so searching for a word that only exists on the whiteboard photo finds the note.

  3. Ask the AI chat to give you the attachment's text as clean Markdown with headings and checkboxes, then paste it into the note body and proofread it against the photo. The chat reads the extracted text, so the request works on photos you attached minutes ago.

  4. Turn the action items into tasks with @due(2026-10-28). They show in the Tasks view and on today's daily note until ticked.

  5. Link the note to its project with [[Sprint 41]] or tag it #bio204. The project note's backlinks panel then lists every board photo and lecture page that mentioned it.

Start with the quick start if you are new to Scribelet.

Frequently asked questions

How do I convert a photo of handwritten notes to text for free?

On iPhone, open the photo and use Live Text to select and copy the words. On Android, use Google Lens. Google Keep's Grab image text and Google Docs (upload the image to Drive, open with Docs) also work without paying. Scribelet's free plan accepts photo attachments and extracts their text with daily AI limits. All of these need proofreading, especially on numbers.

Can Google Keep convert handwriting to text?

Yes, from a photo. Add the image to a Keep note, tap it and choose Grab image text. Keep puts the recognized text into the note under the image. It works best on neat print; cursive and faint pencil produce more errors.

Does handwriting OCR work on cursive?

Partly. General OCR (Live Text, Lens, Keep) reads print far better than cursive. Tools trained on handwriting, such as Pen to Print and Transkribus, do better on cursive and on historical scripts. Expect to proofread every line of cursive whatever you use, and keep the photo as the record.

How accurate is handwritten notes to text conversion?

No vendor publishes reliable figures and your handwriting matters more than the tool. In practice, neat print on lined paper photographed in good light comes out with a few errors per page; rushed notes or cursive can need a correction every line. Numbers, units and proper names are where the errors hide, so check those against the original every time.

How do I turn a whiteboard photo into notes?

Photograph each section of the board square-on before anyone erases it. Run the photo through OCR (or attach it to a Scribelet note, which extracts the text). Then spend five minutes writing what the arrows and groupings meant, because OCR captures words and not relationships. The first example on this page shows the result.

Should I just use a stylus app instead?

If you write on a tablet, yes: Goodnotes, OneNote and Nebo recognize handwriting as you write, and the layout survives. If you write on paper or on whiteboards, those apps do not help; you need the photo route on this page. Many people do both: tablet in meetings, photo for whiteboards and paper.

Photograph the page, keep the original, search the words

Attach the photo to a note in Scribelet and the text is extracted and indexed, with the image kept alongside. Proofread, add the tasks, and the whiteboard outlives the meeting room. Free to start.

Free plan, no credit card. See pricing

Study

Class Notes Template

One note per class session, linked to one note per course. The template captures what the lecture was trying to answer, the key points in your words, the questions you still have, and what is due next. Copy it, or start from the example closest to your class.

Study

Cornell Notes Template

Cornell notes split a page into three jobs: notes during class, cue questions afterwards, and a short summary at the bottom. This version is plain Markdown, so the cues sit above each section of notes instead of in a margin. Copy it, or start from the example closest to your class.

Research

Lab Notebook Examples and Template

A good lab notebook entry lets someone else repeat your work and see what you saw: what you meant to do, what you actually did, the raw numbers, the calculations, and what you will do next. Below are five complete entries from different kinds of labs and the template they share.

Meetings

Meeting Notes Template

One structure that captures the three things people come back for: what was decided, who owns what by when, and the context that made the decision make sense. Copy it as-is, or pick the example closest to your meeting.

Guide

Lecture Recording to Notes: Transcribe First, Then Structure

A recording becomes notes in two passes. First a transcript: your phone, your meeting tool or a lecture app turns speech into text. Then the part that makes them notes: cutting the transcript down into topics, definitions, examples and open questions, checked against the slides. This page covers both passes, the etiquette of recording people, and what the finished notes should look like.

Article

How to remember what you read

From the Scribelet blog

We use cookies for analytics to improve your experience. Learn more