Handwritten notes to text: 9 tools, tested
66.5%
of handwritten words found, Apple Vision
Free on a Mac, 16 lines
0.36 s a note
82 of 100 notes at half the words or more
-
If
swiftis missing, runxcode-select --install. - Save notes.swift below.
- Run
swift notes.swift note.jpg > note.txt. - Check names, dates and numbers against the photo.
notes.swift
// Handwritten notes to text with Apple Vision. Tested on macOS 26.6.
// Run: swift notes.swift note.jpg > note.txt
import Foundation
import Vision
let url = URL(fileURLWithPath: CommandLine.arguments[1])
let req = VNRecognizeTextRequest()
req.recognitionLevel = .accurate // .fast finds almost no handwriting
req.usesLanguageCorrection = true // off cost 10 points here
req.recognitionLanguages = ["en-US"]
// From a file URL, Vision turns sideways phone photos upright for you.
try VNImageRequestHandler(url: url, options: [:]).perform([req])
for line in req.results ?? [] {
print(line.topCandidates(1).first?.string ?? "")
}
The reader ships with macOS, so there is no model to download. On 3 notes the script printed the same text as the bench.
Apple's reader found the most words
| Tool | Found | Right |
|---|---|---|
| Apple Vision, accurate | 66.5% | 69.0% |
| Apple Live Text API | 64.5% | 68.1% |
| Apple documents API (macOS 26) | 64.1% | 67.4% |
| PaddleOCR | 62.4% | 69.3% |
| TrOCR, RapidOCR lines | 56.7% | 56.2% |
| RapidOCR | 56.2% | 59.3% |
| Our Actor | 53.3% | 57.6% |
| EasyOCR | 16.1% | 19.8% |
| Tesseract | 9.3% | 19.2% |
Found is the share of the 5,473 true words that came out. Right is the share of output words that are true.
Apple's three readers took 0.2 to 0.6 s a note. PaddleOCR took about 16 s on the CPU, with other jobs running.
Keep accurate mode and correction on
The fast level trades accuracy for speed. On handwriting it wrote nothing at all for 9 notes.
Sideways photos: Vision copes
Phones save many photos sideways, with a tag that says how to turn them. 19 of the 100 notes have one.
| Run | Words |
|---|---|
| Apple Vision, from the file | 863 |
| Apple Vision, pixels as stored | 836 |
| TrOCR, turned first | 656 |
| TrOCR, pixels as stored | 51 |
From a file, Vision gave the same text as when I passed the tag myself.
Close, not exact
Apple Vision wrote:
Order 4 mile 6
The court has the power to make orders
by way of injunctions and such necessary
Orders for the protection of property on
person pending the determination of an appene
to it even though no Such application was
mack in the court belor.
"Appeal" became "appene" and "made" became "mack". Language correction helps, but it cannot know your names or numbers.
Typed scans to text, no install
Run it yourself. Our Apify Actor turns a scanned PDF into Markdown or text, $0.006 a scanned page. It is built for print.
Or send us the scan. Tack, our AI, does the job for $29, up to 50 pages. Printed text only.
How I tested it
- Notes: 100 test notes from GNHK, camera photos of real handwriting, 5,473 true words. The same notes as Tesseract handwriting and TrOCR.
- Score: true words found in any order, with the scorer of those pages. Right: true words over all output words.
- Apple: macOS 26.6 on Apple Silicon, one note at a time, at the lowest priority. Vision with en-US. Live Text is VisionKit's ImageAnalyzer.
- Time: the median a note, one run, from the call to the text.
- The other rows: the same notes and scorer, run on this Mac. PaddleOCR, RapidOCR, EasyOCR and Tesseract on November 4, TrOCR on November 13. Our Actor: its code on November 17. EasyOCR wrote nothing on 12 notes, which count as 0.
What I did not test
- Tablet note apps such as Goodnotes and Apple Notes.
- Cloud readers such as Google Lens and Microsoft OneNote.
- Apple Vision on an iPhone or iPad.
- Languages other than English, and printed text.
Sources
Questions people ask
What is the best free way to turn handwritten notes into text?
On a Mac, Apple Vision. It found 66.5% of the words on 100 real notes, at no cost. Without a Mac, PaddleOCR found 62.4%.
Can Tesseract read handwritten notes?
Not well. It found 9.3% of the words on the same 100 notes. It is built for print.
Can I turn handwriting into text without code?
On a Mac, open the photo in Photos and select the text with Live Text, then copy it. The Live Text API found 64.5% of the words here. I tested the API, not the menus.