One wrong word on a scan used to mean retyping the page. Mine changes

0.9% of the pixels
See what I built ↓

Forty seconds. One word changed, nothing else.

How it changes only the word.

The typeface comes from the pixels

A scan stores no font name. It draws the word in every typeface on the machine, compares the pixels, and fits the size, blur and stroke weight until its redraw matches what is on the page.

Read from the scan
Typeface
Size
Blur
Stroke
✓ Solved, not guessed
The word “recieved”Test page · real run
  • FaceGeorgia Regular
  • Size30 px
  • Blurσ 0.8
  • StrokeGamma 1.1

Less than 1% of the page changes

Only the letters that differ are redrawn. The watermark, the rules and the grain around them keep their original pixels.

Pixels changed2,609
Of the page0.9%
Elsewhere0
The page, as 220 squares2 changed
UntouchedChanged

The paper under the word is rebuilt

It never paints a box over the word. Only the ink of the old letters is removed, and the grain is carried in from a clean strip beside it.

✕ Never used The usual way
a filled white box

Erases the watermark and the grain

✓ First Ink mask
only the old letters

Paper rebuilt under them

✓ Then Grain transfer
from a clean strip

Texture matches the page

It runs on your own machine

Typeface matching, the paper rebuild and the edit all happen on the laptop. The document does not have to leave it.

Your laptop Local
Typeface matchingLocal
Paper rebuildLocal
Images and PDFsLocal
A cloud editorNot used

It grades its own edit

Every edit is measured against untouched text on the same page, on six measures. Coming out cleaner than the scan around it counts as a fail.

Edge sharpness100.4 vs 111.3
Noise1.20 vs 0.88
Ink darkness49 vs 41
Paper level250.3 vs 253.3
Stroke weight3 vs 3
Colour fringing0.51 vs 0.15
100out of 100 edit vs the page · real run on the test page

What does your teamstill do by hand?

Send me the task that eats an afternoon every week.
I'll tell you what it would take to build the tool.

Send me the problem → All tools