InSnap Blog

Why OCR Returns the Wrong Word Instead of an Error

Arthur Zhang — Wed, 28 Oct 2026 18:50:03 GMT

Copying text out of a screenshot usually works, and that is the problem. When it fails, it rarely fails loudly. There is no red underline, no "could not recognise text", no empty box. There is a sentence that reads perfectly well — except for one character, and that character changes the meaning.

This is why people paste an order number that is off by a digit, a config key with a wrong letter, or a sentence in a document with a character that was never there. The recognition did not crash. It guessed confidently.

Recognition is a classifier, not a parser

It helps to know what the engine actually does. Text recognition splits the image into lines, then into character-shaped regions, then runs each region through a classifier that returns the most likely character. That classifier never says "I don't know". It always returns the closest match it can find.

At good input quality, that closest match is right almost everywhere. At poor quality, the closest match is a different character that happens to look similar: an a that could be an o, a 1 that could be an l, or one CJK character that could easily be mistaken for a neighbour with a shared component. The engine has no idea it is wrong.

What makes the input "poor"

Four things matter, in roughly this order:

Text height in pixels. Smaller text has fewer pixels to classify. This is the big one.
The character set. Chinese, Japanese and Korean characters are dense; a few lost pixels change both the shape and the identity of the character. Latin letters are simpler and survive longer.
Anti-aliasing and scaling. Every resize blends the edges of every stroke. One resize is fine; three, stacked, lose the shape.
Contrast and background. Grey-on-white and light-on-dark both reduce edge strength. So do gradients and photos behind the text.

Most of the failure reports come from the first one, and the fix is usually to capture more pixels, not to try harder.

InSnap has no separate accuracy setting. It works on the pixels you select: a 1x screenshot of a dense UI already has each line at the height the screen drew it, and a 2x Retina capture has double that. The difference between the two, for a given paragraph of text, is often the difference between correct and plausible-but-wrong.

Where the practical line sits

To see the boundary rather than guess at it, it is worth building a small, repeatable test: render the same sentence at several text heights, run the recognition on each, and compare the output to the input character by character. That is exactly what the two images below do, using real output from a local engine, not made-up examples.

The Chinese line and the English line were rendered at five heights and recognised fresh each time. Red marks the characters that differ from the original.

Two things come out of it. First, the line is correct until it is not: there is no gradual degradation, just a size below which the wrong character appears. Second, the two scripts break at different points. Latin text slipped earlier here, while Chinese held on longer, which is the opposite of what most people assume from "Chinese is harder".

When the same sentence is recognised at even smaller sizes, the output becomes a different kind of wrong. Not empty, not gibberish, but a real sentence with real, wrong characters:

That second image is the useful one. It shows what a wrong result actually looks like: not obvious enough to notice, not subtle enough to be nothing.

How to catch it before it reaches a document

The recognition cannot tell you it is unsure, so you have to check in places where a wrong character is expensive.

Check anything with numbers or identifiers. Order IDs, licence keys, version strings, invoice totals. A single wrong digit is invisible in context and fatal downstream.
Check the length. A two-character identifier that came back as three, or the reverse, is a reliable signal that something went wrong.
Check unusual characters. Pipes, spaces, non-breaking spaces, Cyrillic versus Latin confusion. If the text is code, pasting into a monospace view makes these obvious.
Check the edges. Characters at the start or end of a line are the first to be cut off during detection. If a line ends abruptly, look at the last one or two characters.
Recapture at a larger size, do not upscale. Upscaling a blurry line does not create detail, and it rarely fixes a wrong character. Recapturing the same area at 2x, or zooming the source before capturing, does.

The practical corollary is that a screenshot taken with a wide selection at good resolution is easier to recognise than a tiny one cropped tightly. Cropping reduces noise, but shrinking reduces information, and information is what the classifier eats.

What you can rely on

The honest summary:

Situation	Usually fine	Usually not
English and Latin-script text at normal UI size	Yes	—
Chinese at normal UI size	Yes	—
Text under roughly 10 px tall	—	No
Japanese kana, Korean, Russian	—	No
Text over a photo or gradient	Sometimes	Sometimes
Heavily resized screenshots	—	No

InSnap's local recognition handles Chinese, English and Latin-script languages well, including accented characters. It is not reliable for Japanese kana, Korean or Russian. That is a property of the bundled model, not a setting, so the honest move for those scripts is to verify anything you copy.

FAQ

Does the text recogniser tell me when it is unsure?

No. A character classifier always returns a best guess, and the interface has no separate confidence indicator for copying. The safe assumption is that any character it returns is the most likely one, not a guaranteed one.

Can I fix a wrong character by running it again?

Not usually, because the input has not changed and the model is deterministic on the same pixels. The fix is to change the input: recapture the same region at a larger size, or zoom the source before capturing, then recognise again.

Why does it work on one screenshot and fail on another of the same page?

Most often text size in pixels, which depends on the display scaling and on whether the capture was at 1x or 2x. The same page at a different zoom or on a different display can cross the size boundary.

Does it read text inside images and photos?

Only if the text is large and clear over a plain background. Text over a photo, a gradient or a complex illustration is where wrong characters most often appear.

Do I need an internet connection for this to work?

No. Recognition runs locally, including the copy step. Nothing is uploaded, and the accuracy is the same offline.

How to Copy Text From a Screenshot or Image on Mac and Windows

Arthur Zhang — Tue, 27 Oct 2026 16:45:30 GMT

You have a screenshot, and you need the words inside it. OCR (optical character recognition) turns those pixels back into selectable text. On a Mac, open the image in Preview or Quick Look, drag across the words and press Command-C. On Windows 11, take a snip and use Text actions in Snipping Tool, or press Windows key + Shift + T with PowerToys installed. If you have not captured anything yet, InSnap does both at once: drag a region, click OCR on the capture toolbar, and copy.

Built-in tools are free. InSnap is faster when the text is already on screen.

On a Mac: Live Text

Live Text has shipped with macOS since Monterey and works in Preview, Quick Look, Photos and Safari.

Preview: open the image, hover until the pointer changes, drag to select, and press Command-C. If nothing selects, use Tools › Text Selection.
Quick Look: select the file in Finder, press Space, drag over the text, and press Command-C. The Live Text icon offers Copy All.
Photos: open the photo and select text the same way.

Recognition runs on the Mac; available languages depend on your macOS version and region.

On Windows: Snipping Tool and PowerToys

Snipping Tool (Windows 11): press Windows key + Shift + S, open the snip, click Text actions, then select words or use Copy all text. The Windows 10 Snipping Tool lacks this.
PowerToys Text Extractor: a free Microsoft utility for Windows 10 (version 2004 or newer) and Windows 11. Enable it, press Windows key + Shift + T, and drag a box around the text. It uses the OCR language packs in Windows.

Copy text with InSnap, on Mac and Windows

InSnap is a screenshot and annotation tool for macOS 14.4+ and Windows 10+. Its OCR is free, local, and works the same on both platforms.

Press Command-1 (Ctrl+1) and drag a box around the text.
Click OCR on the toolbar; recognition runs in the background while the capture stays on screen.
Select text as in a document: drag across characters, click a line to select it whole, or Shift-click to add more lines. The panel beside the selection lists the same lines.
Press Command-C (Ctrl+C) to copy the selection. With nothing selected, copying takes all the text, and the panel has buttons to copy the selection or everything.

Get better recognition results

Crop tightly. Select the text and little else.
Capture at full resolution. A real screenshot beats a stretched copy; upscaling adds no detail.
Keep text horizontal. A photo taken at an angle introduces skew.
Avoid glare, shadows and moiré from photographing a glossy screen.
Use readable text sizes. Very small text is where errors start.

Which languages are recognised

InSnap reads Chinese (Simplified and Traditional), English, and Latin-script languages including German, French, Spanish, Portuguese and Italian, accents included.

Japanese kanji are often recognised while kana frequently come out wrong. Korean and Russian are not reliably recognised yet. Lines mixing Chinese and English usually work; scripts the model cannot read will not.

Your image stays on your device

Recognition uses a bundled model and runs on your Mac or PC. Nothing is uploaded, no account is needed, and it works offline, which matters when a screenshot holds client names or internal tickets.

Built-in tools vs InSnap

	Built-in tools	InSnap
macOS	Live Text in Preview, Quick Look and Photos	OCR on the capture toolbar
Windows	Snipping Tool text actions (Windows 11); PowerToys (Windows 10 and 11)	OCR on the capture toolbar
Select text on the image	Yes	Yes, on the image or in the panel
Languages	Depends on the OS and language packs	Chinese, English and Latin-script languages
Cost	Included with the OS; PowerToys is free	Free

Troubleshooting

Nothing is recognised. Check that the selection contains text and is not a tiny box. Logos, handwriting and text drawn into an illustration are out of scope. Use a real screenshot, not a scaled image.

Characters come out wrong. Lookalikes such as O and 0, l and 1, or rn and m, plus stylised fonts, low contrast and small sizes cause most errors. Proofread codes and amounts before pasting.

Mixed languages. Select only the lines you need, with one script per selection where possible.

FAQ

How do I copy text from a screenshot without extra software?

On a Mac, use Live Text in Preview or Quick Look. On Windows 11, take a snip and use Text actions in Snipping Tool.

Is OCR free in InSnap?

Yes, with no Pro licence or account required.

Which languages can InSnap's OCR read?

Chinese (Simplified and Traditional), English, and Latin-script languages such as German, French, Spanish, Portuguese and Italian. Japanese kana, Korean and Russian are not reliably recognised.

Does InSnap upload my screenshot for OCR?

No. Recognition runs on your computer, and the image never leaves your device.

Can I copy only part of the recognised text?

Yes. Drag over words on the image, click a line, or select in the panel, then copy. With nothing selected, everything is copied.

How do I extract text from an image on Windows 10?

Snipping Tool text actions are only in Windows 11. On Windows 10, use PowerToys Text Extractor, or capture with InSnap and press OCR.

How to Make a Step-by-Step Guide by Recording Your Screen

Arthur Zhang — Tue, 27 Oct 2026 07:48:19 GMT

A step-by-step guide is a series of numbered screenshots with short instructions: what to click, what to type, and what should appear next. The fastest way to build one is to do the task once while a guide recorder runs. Each click becomes a step with a screenshot, a marker on the target, and a title taken from the button or field name. You edit the result like a document instead of writing the whole thing by hand.

Guide recording in InSnap is a Pro feature. Start it from the menu bar or system tray, do the task, and stop; the guide opens in an editor where you can rename steps, reorder or hide them, annotate screenshots, blur private details, and export. Everything is processed on your computer; files leave it only if you share them through the optional Cloud subscription.

What makes a guide useful

A useful guide lets a reader follow it without guessing. That depends less on the number of screenshots than on how each step is written.

One action per step. Opening a menu and choosing an item are two steps.
Titles written as instructions. "Click Save" says what to do; "The Save dialog" only describes what you saw.
A visible result. Show enough screen to confirm the reader is in the right place.
Crops with context. Remove unrelated windows, but keep the control being clicked and the area that changes.

Record the task once

Start a guide recording from the InSnap icon in the menu bar or system tray, or with a hotkey you set in Settings › Hotkeys. Choose the whole screen, one app, or a region.
Do the task. Each click, or a key you choose, adds a step. Double-clicks are merged and drags are ignored.
Stop the recording. The guide opens in the editor.

Close unrelated windows and notifications first; the screenshots show whatever is on screen.

Edit the steps

Rename titles so they say what the reader should do, and describe anything that needs a reason.
Reorder steps by dragging, hide one without deleting it, or add text-only steps.
Add sub-steps, tips, and warnings to separate the main path from details.
Annotate with arrows, boxes, or mosaic, and crop each screenshot to the part that matters.

Hide or blur private details

Recordings pick up more than the task. Check every screenshot for names, email addresses, account numbers, internal URLs, and customer data.

Hide a step only some readers need instead of breaking the numbering.
Blur or mosaic the sensitive area. The blur can be applied across all steps at once.
Recheck the first and last steps, which most often show an account or a file path.

Export in the right format

Pick the format that matches where readers will open it.

Format	Best for
PDF	A fixed document for printing and sharing
Word	A team that edits or adapts the text
PowerPoint	Presenting or splitting into slides
HTML	A single page for an intranet or ticket
Markdown	A docs folder or wiki
SCORM 1.2	A learning system
JSON	Another tool that reads the steps

Add a cover page if you like. Guides are part of Pro, a one-time {{proPrice}} purchase with no subscription.

Quality checklist before you share

Run through the guide as if you were the reader.

Start from a clean state, so step 1 matches what the reader sees.
Confirm each title names a control that is visible in its screenshot.
Check that numbering stays continuous after hidden steps.
Zoom in on each blurred area and make sure the text underneath cannot be read.
Open the exported file, not just the editor.
Ask someone unfamiliar with the task to follow it.

When a screen recording works better

Some subjects are easier to show than to describe. A recording is better when the point is motion: dragging an item, drawing, an animation, or a transition between screens. It also helps when timing matters, such as waiting for a build or a page to load, and when the path depends on what appears on screen during troubleshooting. A guide is easier to skim, search, print, and follow at your own pace, and easier to update when a step changes.

Basic screen recording to MP4 is free in InSnap; GIF, auto zoom, camera picture-in-picture, and video editing are Pro.

FAQ

How do I create a how-to guide quickly?

Record yourself doing the task once. Each click becomes an annotated step that you can rename, reorder, and export.

Can I export a guide to Word or PDF?

Yes. InSnap exports PDF, Word, PowerPoint, HTML, Markdown, SCORM 1.2 and JSON, with an optional cover page.

Does recording a guide upload my screen?

No. Guides are recorded, stored, and processed on your computer. Files are uploaded only if you share them with the optional InSnap Cloud.

Can I blur sensitive information in every step at once?

Yes. A blur or mosaic applied in the editor can be applied across all steps, so you do not have to fix each screenshot separately.

Is guide recording free?

No. Guide recording and export are a Pro feature. Basic screenshot, annotation, pinning, OCR, and basic screen recording remain free.

Disclosure: I build InSnap. The workflow above works with other tools too.

Originally published at https://insnap.app/blog/make-a-step-by-step-guide-from-screen-recording/

How to Blur or Hide Sensitive Info in a Screenshot Safely

Arthur Zhang — Tue, 27 Oct 2026 07:43:27 GMT

To blur or hide sensitive information in a screenshot, cover it before the image leaves your computer. In InSnap, choose Mosaic and raise Strength until the text is unreadable, or draw a Filled shape in a solid color over the data. Then check the whole frame, not just the spot you focused on.

A demo account name can take a strong mosaic. Passwords, API tokens and customer records should not be visible at all: use a solid cover or crop them out.

Mosaic, blur or a solid cover?

Three treatments look similar but behave differently:

Mosaic (pixelation) replaces the area with large blocks and reads as "redacted".
Blur smooths the area. It looks softer but keeps more of the original signal.
A solid cover is a filled shape in one opaque color; the pixels underneath are replaced.

Mosaic and blur transform the area; the original information is still encoded in the result. With small blocks, light blur, short text and a predictable font, part of the string can sometimes be recovered: deconvolution for blur, tested guesses for pixelation. Sharp interface text is the easiest case. A solid fill avoids that, and cropping removes the area from the file entirely.

What is on screen	Recommended treatment
Passwords, 2FA codes, API keys, tokens	Do not capture; solid fill or crop
Card numbers, CVV, bank details, IDs	Solid fill or crop; never mosaic or blur
Customer names, emails, addresses, order data	Solid fill; strong mosaic only for invented demo data
Faces and avatars	Mosaic, blur, or crop the person out
Internal URLs, host names, account IDs	Strong mosaic, or solid fill if sensitive
Tabs, bookmarks, notifications, menu bar	Crop the strip, or cover it

Hiding data with InSnap

Press the capture hotkey (Command-1 on Mac, Ctrl+1 on Windows) and drag a box.
For general redaction, choose Mosaic, pick Mosaic or Blur, and raise Strength until the text is gone. Drag over each detail; boxes can be moved and resized.
For a secret, choose Shape, draw a rectangle, set the style to Filled and pick a solid color. The fill is opaque; the text underneath is not in the export.
If the sensitive part is not needed, crop it: take a smaller selection, or crop in the image editor.
Copy, save or pin the result; the hiding is applied to the exported pixels.

Smart Erase removes a stray name or order number and fills the gap from the background. It works on flat and gradient backgrounds, less well over photos. It is a cleanup tool, not a redaction guarantee: for customer data and secrets, use a solid cover. Part of Pro.

Watermark adds a visible label, useful for marking a copy as draft or internal. It hides nothing, so treat it as a label, not protection. None of these tools stops a determined reader; they reduce what is readable in the shared file.

Blur in all steps when you record a guide

A guide captures the same window many times, so the customer name in step 2 is usually still in step 3 and the exported PDF.

In the guide editor, draw the mosaic once, then click Blur in all steps above the tools. InSnap copies it to every step, scaled and at the same relative position. That assumes the layout stays put: check steps captured after a scroll or in a different dialog. The guide editor and its redaction tools are part of Pro.

A checklist before you send

Read the result at 100%. Confirm the text under each cover is unreadable, not just blurry.
Scan the whole frame. Tabs, bookmarks, notifications, the menu bar, the Dock, window titles, URLs, anything on a second display.
Check the file name and any title above the image. invoice-acme-2026.png can leak more than the picture.
Check the edge of every redaction. A box that was moved can leave a letter outside it.
Check every image in a batch, and decide what happens to the local original. Hiding data in the shared file does not delete it from your disk.

Never leave these visible: passwords and 2FA codes, API keys, tokens and private keys, card numbers and CVVs, government IDs, customer names, emails, addresses and order data, faces of people who have not agreed to appear, private messages, and internal URLs or host names.

FAQ

Can blurring a password make it safe to share?

No. Use an opaque filled shape or crop it out. If it was in a file you already sent, change it.

What is the difference between mosaic and blur?

Mosaic replaces the area with blocks; blur smooths it. A stronger setting hides more, but neither deletes the original pixels the way a solid fill does.

Does Smart Erase delete the data it removes?

It replaces the pixels inside the box with content from the background, and never changes pixels outside it. Flat backgrounds come out clean; busy photos are a guess, so do not rely on it for secrets.

Can someone undo the mosaic in an image saved by InSnap?

Not by opening the file: the exported image holds the redacted pixels, not an original layer. Weak settings on large, sharp text can still be guessed, so use a strong setting or a solid cover.

Disclosure: I build InSnap. The workflow above works with other tools too.

Originally published at https://insnap.app/blog/blur-or-hide-sensitive-info-in-screenshots/