Extracting text from an image requires OCR, and a PDF may not need OCR at all. That distinction is the single most useful thing to know before you pick a tool. Many PDFs already contain a selectable text layer; those can be read locally in a browser with no recognition step. A photograph of a page or a scanned document contains only pixels, and turning those into characters needs a recognition model. UsefulCrate does not currently offer OCR, so this guide separates the two cases and gives you a free, local path for each.
Case 1: the PDF already has a text layer
Most PDFs created by Word, a browser's print dialog, an invoicing system or an export button contain the actual characters along with instructions about where to draw them. When you can drag across a line in a PDF viewer and the text highlights, that is a text layer. No recognition is needed; the words are simply packaged in a format that is awkward to edit.
The PDF to Word tool reads that layer locally and rebuilds it as an editable .docx, carrying the document's pictures over as well. It preserves paragraphs, headings and reading order, and it works entirely in the browser tab — useful when the document contains client names or commercial terms you would rather not upload to a free converter.
How to tell in five seconds
Open the PDF and try to select a sentence with your mouse. If characters highlight individually, you have a text layer. If the whole page is one rectangle or nothing highlights at all, it is an image. You can also watch the panel after conversion: a browser tool will show a sample of the extracted text, which is the fastest confirmation that the layer was actually there.
Case 2: the page is really an image
Phone photos of a whiteboard, screenshots saved as PNG, faxed forms and flatbed scans made without OCR contain no characters at all — just a grid of pixels. Reading them requires a recognition model that identifies letter shapes, groups them into lines and words, and guesses the language. That is OCR.
| Input | Contains | Local method |
|---|---|---|
| Exported PDF from Word | Real text | Extract the text layer |
| Digital invoice PDF | Usually real text | Extract the text layer |
| Scanned PDF, no selection | Pixels only | OCR |
| Photo of a page | Pixels only | OCR |
A browser-based conversion of a scanned PDF will place the page into the output document as a picture and say plainly that the text is not editable. That is the honest result: an image cannot become editable text without recognition, and pretending otherwise would produce a document full of empty paragraphs.
Free, no-upload OCR options
Use the tools already on your device
iOS and macOS have Live Text: point the camera at a page or open a photo, and the operating system lets you select and copy the words. Android Google Lens does the same, and Windows PowerToys includes a Text Extractor that copies text from any region of the screen. These are free, run on your device, and need no signup. For a handful of pages they are the best option available.
Use a local OCR engine
For a batch of scans, a desktop engine such as Tesseract produces a searchable PDF or a plain text file entirely offline. It is a command-line tool with a learning curve, but it is genuinely free and keeps every page on your machine. Some browser pages wrap an OCR model compiled to WebAssembly, which gives the same privacy property without installing anything; check the Network panel to confirm no image is uploaded.
Be careful with free online OCR
A service that accepts an image and returns text is processing it somewhere. That is a reasonable trade for a public sign or a printed manual, and a poor one for a contract, a medical form or an identity document. If you must use one, remove unrelated parts of the image first and read the retention policy.
Improve the input before recognising anything
OCR accuracy is decided before the model runs. A clean, flat, well-lit page beats any amount of post-editing. Four things make the biggest difference:
- Flatten and straighten. Shoot from directly above, with the whole page in frame and no perspective skew. A tilted baseline makes line segmentation much harder.
- Increase effective resolution. Small text needs enough pixels per character. Cropping out the empty desk around a document often helps more than zooming.
- Fix contrast and shadows. Even lighting and dark text on a light background are what recognition models are trained on. A phone shadow across half the page will lose words.
- Use a lossless format. A heavily compressed JPEG smears the edges of thin glyphs. The Image Converter can produce a PNG from a JPEG source, though it cannot restore detail that compression already removed — reshooting is better.
For a set of page photos, resizing them to a consistent width with the Image Resizer and assembling them into one ordered file with JPG to PDF makes a far more manageable input for any OCR engine than a folder of mismatched images.
After extraction, verify rather than trust
OCR is a guess, and confident-looking guesses are the dangerous kind. Numbers, names, dates and currency amounts deserve a line-by-line check against the original image. A single misread digit in an invoice total or a dosage is not a typo you will notice while skimming.
Two local checks help. Paste the extracted text into the Word Counter and compare its length with what you expect from the page; a suspiciously short result usually means a column or a page was skipped. Then use the Text Cleaner to remove the stray blank lines and broken words that recognition commonly leaves behind before you paste the text into a document.
The known limits nobody can engineer around
Handwriting, low-resolution faxes, heavily stylised fonts, text over photographs and dense tables all produce unreliable output. A PDF that omits its character mapping presents a different problem entirely: the glyphs are there but the information about which letter each one represents was never written into the file. No converter can recover that, including desktop software; the source document or a re-export is required. This is worth knowing before spending an afternoon trying different tools on the same file.
Common questions
Can the PDF to Word tool on this site do OCR?
No. It extracts an existing text layer and carries images across as pictures. A scanned page comes through as an image with a note that its text is not editable. OCR needs a recognition model, which we do not currently ship.
Why did my conversion produce garbled characters?
If the source is a PDF, check whether it has a text layer at all and whether the font carries a character mapping. Some older files omit that mapping, and the resulting nonsense is not a converter bug — the information is missing from the file. Re-exporting from the original application fixes it.
Is there a completely free way to OCR a hundred pages?
Yes, a local desktop engine such as Tesseract, or a WebAssembly OCR page that processes on your device. Both avoid per-page fees and uploads, at the cost of setup time or a model download. For a one-off document, your phone's built-in Live Text is faster.
Tools on this site
PDF to Word extracts text from PDFs that already have a text layer. Word Counter gives a quick sanity check on extracted length. Text Cleaner tidies the blank lines and broken words that recognition leaves behind.