How browser OCR reads text from photos and screenshots
Local optical character recognition runs through a self-hosted Tesseract.js WebAssembly worker loaded directly from this site’s static assets. Every image is decoded in browser memory, normalized onto a high-contrast white canvas, and scanned for English character shapes without sending a single byte over the network.
If your screenshot contains structured headings, bullet lists, or multi-column grids that you want to preserve as syntax, you can reconstruct headings and pipe tables from screenshots instead of extracting flat plain text.
Automatic upscaling below 1,600 px and 4,000 px limits
Small UI captures and cropped chat screenshots often have character heights under 14 px, which causes raw OCR models to confuse i, l, and 1. Before recognition begins, prepareOcrCanvas checks the source width and enlarges any image narrower than 1,600 px by up to 3x.
To prevent browser memory crashes on giant camera files, scaling uses MDN Web Docs CanvasRenderingContext2D drawImage with high smoothing quality while enforcing a strict 4,000 px maximum side and a 16,777,216 px total area ceiling. Images above those bounds are scaled down proportionally so recognition completes reliably on mobile and desktop tabs.
White canvas backing for transparent PNGs and SVGs
Screenshots of dark icons or transparent PNG badges often carry an alpha channel where transparent pixels default to black when flattened naively. The preprocessor fills the working canvas with solid #fff white before drawing the decoded image on top.
This white backing guarantees crisp contrast for dark text on transparent PNG or SVG inputs. Once Tesseract finishes reading the canvas, the temporary canvas dimensions are zeroed out immediately to release graphics memory.
Pasting screenshots, batching files, and supported formats
You can queue one image or an entire folder of receipts, slides, and phone photos at the same time. Files are processed sequentially in the order they appear so your browser tab stays responsive while reading large batches.
When your source document is already a multi-page PDF rather than a folder of images, pull text from multi-page PDF documents directly without exporting individual page screenshots first.
Instant clipboard capture with Ctrl+V and Cmd+V
A global listener attached to the MDN Web Docs Clipboard paste event inspects clipboardData.files whenever you press Ctrl+V or Cmd+V outside a text field. Copied PNG screenshots are wrapped with a Pasted image <n>.png label and queued immediately.
This clipboard workflow skips the extra step of saving a temporary screenshot to your desktop before uploading. You can paste multiple screenshots in a row to build a single combined transcript.
Sequential queue decoding for HEIC, SVG, and raster photos
The drop zone accepts standard raster formats alongside iPhone HEIC/HEIF photos and SVG graphics.
| Input or control | Behavior in tool | Key numeric limit or rule | Best use case |
|---|---|---|---|
| Choose images / drop | Queues multiple image files | Accepts image/*, .heic, .heif, .svg |
Receipts, book pages, scans |
| Clipboard paste | Reads Ctrl+V or Cmd+V images |
Auto-names image.png sequentially |
Snipped UI dialogs and errors |
| Keep line breaks as in the picture | Preserves or unwraps line breaks | Collapses 3+ newlines to 2 | Lists vs flowing paragraphs |
| Copy text / Copy all text | Copies active output to clipboard | Joins batch items with filename headers | Quick pasting into notes |
| Save .txt | Downloads UTF-8 plain text | Saves as text-from-images.txt |
Archiving batch transcripts |
Controlling line breaks and fixing low-confidence OCR
Raw OCR engines break lines at the right edge of every column or paragraph just as they appear in the picture. Depending on your task, such as copying a narrow column of prose or a structured receipt, you can toggle line preservation in real time without re-running OCR.
If a scanned sign, slide header, or form label comes out in all uppercase letters, you can fix all-caps OCR output into sentence or title case right after copying.
When to uncheck Keep line breaks as in the picture
By default, Keep line breaks as in the picture stays checked so tables, code lines, and postal addresses retain their visual line structure. Trailing spaces and tabs before line endings are trimmed, and runs of 3 or more blank lines are collapsed to 2.
Unchecking Keep line breaks as in the picture runs a pass that replaces single line breaks with spaces while keeping blank-line paragraph breaks intact. Standard UTF-8 output respects word and line boundaries defined in Unicode Standard Annex #29 on Text Segmentation, making unwrapped paragraphs paste cleanly into word processors.
Interpreting the confidence percentage score below 60
Tesseract calculates an overall recognition confidence percentage score from 0 to 100 for every processed image. Whenever an image scores below 60, a notice appears beneath the textarea reminding you to verify the extracted text carefully.
Troubleshooting blurry photos and unreadable fonts usually points back to motion blur, uneven desk shadows, low print contrast, or photographing a glossy monitor at an angle. Cropping tightly around the text block or capturing a direct digital screenshot almost always raises confidence above 60.
Why local Tesseract WASM keeps sensitive text private
Screenshots frequently contain private API keys, medical receipts, bank statements, or internal company dashboards that should never touch a third-party OCR server. Running Tesseract WebAssembly locally ensures those pixels stay strictly inside your browser’s sandboxed memory.
Once you finish extracting a long passage, you can paste it into a word and character count checker to verify length limits for forms or summaries.
Self-hosted worker assets and bitmap memory cleanup
The tool decodes each image file using the MDN Web Docs createImageBitmap API and closes the bitmap immediately in a finally block once OCR completes. All Tesseract worker scripts and English traineddata files load from /vendor/tesseract/ on this origin with zero external API calls.
Tip: Click Clear all after finishing a sensitive batch. Clearing the queue revokes every preview object URL, empties the in-memory file map, and terminates the Tesseract worker thread to free RAM.
Exporting single copies vs combined text-from-images.txt
For a single image, the Copy text button copies the exact contents of the textarea, honoring your Keep line breaks as in the picture setting. When multiple images are loaded, Copy all text and Save .txt prepend a ── <filename> ── header above each image’s block and separate files with a blank line.
Downloading via Save .txt writes a clean text-from-images.txt file directly from browser memory. Because the line-break toggle is reactive, you can switch modes and re-export instantly without waiting for OCR to run a second time.