JPG to Word

Drop a photo or screenshot here

Or paste one with Ctrl+V (⌘+V on a Mac). It's read on this device. The photo is never uploaded.

Language on the page

The English model is about 5 MB and downloads once, then stays cached.

Printed type only. Handwriting almost never comes back right, whatever tool you use. The photo stays on this device. The OCR engine and its language model download once; your photo is never part of that.

Someone hands you a printed page. A contract, a recipe, a page of a book, a letter that predates email. You need the words in a document you can edit, and retyping it is twenty minutes you'd rather spend on anything else. So you photograph it, and then you find out that the hard part isn't the photograph — it's what most tools do next. Many of them simply paste the picture into a Word file and call that a conversion: the document opens, there's your photo, and not one word of it is editable. The ones that do run real OCR usually want the file uploaded first, which means a page you may not own the right to share sits on someone else's server for a retention window you're asked to take on trust. This page does the recognition in the tab you're looking at, using Tesseract — the same open-source OCR engine that sits underneath a lot of paid products — compiled to run in the browser. The photo never leaves your machine. And it does the part almost nothing else bothers with: OCR reads a page line by visual line, so its raw output has a break at the end of every printed line. Drop that straight into Word and the document looks fine until you edit a sentence, at which point nothing reflows, because every line is its own paragraph. This tool joins the wrapped lines back into real paragraphs, mends words the typesetter split across a line break, and hands you a document that behaves like one you typed.

Built by Bob Article by Lace QA by Ben Shipped September 21, 2026

How to use

1
Check the language on the page before you start. It follows the site's language — Spanish on the Spanish page, Polish on the Polish one — so change it if the page you photographed is in something else. The model downloads once — about 5 MB for English — and the browser caches it after that.
2
Drop a JPG, PNG or WebP onto the box, use Choose an image, or just paste a screenshot with Ctrl+V (⌘+V on a Mac) — a paste starts the read straight away. Up to 25 MB and 50 megapixels; a straight-on, well-lit photo reads far better than a bigger file does.
3
Wait through the read. OCR is real computation and it's happening on your device, so a dense page takes a few seconds — longer on a phone than a laptop.
4
Check the notes under the result. They tell you when confidence was low enough that the text needs a proofread, or when the photo was too small for the type to resolve.
5
Choose Letter or A4, decide whether the original photo goes in under the text, and press Download. receipt.jpg comes back as receipt.docx.

📝 PDF to Word 🔍 OCR PDF 🖼️ JPG to PDF 📄 Word to PDF 📋 PDF Extract Text

Frequently asked questions

Is my photo uploaded anywhere?

No. The image is decoded and read inside this tab, and the .docx is written on your own machine — there's no server in the path and nothing to delete afterwards. What does come over the network is the OCR engine itself: Tesseract's worker script, its WebAssembly core, and the language model for the page you picked — about 5 MB for English. Those are fetched from a public CDN the first time you use a language and cached by your browser after that, and they're the same files for everyone; none of them carries your photo. If you want to check, open your browser's network tab while you convert: you'll see the engine and the model come down, and nothing going up.

Does this actually read the text, or just paste my picture into Word?

It reads it. That distinction is worth being blunt about, because a lot of tools with this exact name do the second thing — they wrap your JPG in a .docx, which produces a file you can open in Word and can't edit a single word of. This runs optical character recognition over the pixels and gives you typed text in editable paragraphs. You can optionally keep the original photo underneath, which is genuinely useful for proofreading against, but the text above it is real text.

Why do other converters give me one paragraph per line?

Because that's exactly what OCR hands them. Tesseract reports a page the way it sees it — one line of output for each line of print — and a converter that passes that straight through gives you a document where every visual line is its own paragraph. It looks correct on screen. Then you delete a word in the third line and nothing reflows, because there's a hard paragraph break at the end of it. Fixing that by hand on a full page is worse than retyping. This tool joins wrapped lines back together, splits paragraphs on blank lines and on short sentence-ending lines (which is what the end of a printed paragraph looks like), and rejoins words that were hyphenated across a line break, so "infor-" and "mation" come back as one word.

How accurate is it, and what makes it better?

On a flat, well-lit, straight-on photo of ordinary printed type, Tesseract is very good — the sort of accuracy where you're fixing a handful of characters on a page, not rewriting it. What hurts it, in order: angle, focus, shadow, and unusual fonts. A page photographed at a tilt is the single most common cause of bad output, and it costs nothing to fix. Resolution matters up to a point and then stops: around 300 DPI, roughly 2,500 pixels across a letter-size page, is where it becomes reliable, and past about 4,000 pixels on the long side you're just making the file bigger. When confidence comes back low, the tool says so rather than handing you a clean-looking document full of quiet mistakes.

Will it read handwriting?

Almost certainly not, and no browser tool will. Tesseract is trained on machine-printed type; handwriting is a different problem that needs a different class of model, and even the paid services that claim it are unreliable on anything but very neat block capitals. If your page is handwritten, the honest answer is that typing it yourself will be faster than correcting what any of these tools give back.

What about columns, tables, and the layout?

You get the text in reading order as paragraphs — not a reconstruction of the page. Tables come back as lines of text rather than as Word tables, two-column layouts can interleave, and typeface, size, bold and italic are not carried over. That's a deliberate limit rather than an oversight: reliable layout recovery needs a full document-structure model, and a tool that half-guesses at it produces a mess that takes longer to clean up than plain text does. If the layout is the point, turn on the option to include the original photo, and rebuild the structure around the text you've been given.

Which languages does it handle?

English, Spanish, French, German, Portuguese, Italian, Russian and Polish. The picker starts on the language of the page you're reading this on, so the Polish page reads ą, ę, ł and ż with the Polish model rather than guessing at them with the English one. Check it before you convert — it's not a preference, it's which model gets loaded, and running an English model over a French page produces exactly the sort of confidently wrong output you'd expect. If the page is bilingual, run it once per language and keep the better half of each.

Does it take PNG screenshots, HEIC, or a PDF?

JPG, PNG and WebP all work, whatever the name says — the name is just what people search for. A screenshot doesn't even need saving: copy it and paste it anywhere on the page, and the read starts on its own. An iPhone photo saved as HEIC works only in browsers that can open HEIC, which today mostly means Safari; anywhere else, run it through HEIC to JPG first. For a scanned PDF — which is the same problem with a different wrapper — OCR PDF does the whole job in one step, including multi-page documents. And if the PDF already has selectable text in it, you don't want OCR at all: PDF to Word pulls the real characters straight out, which is always more accurate than reading a picture of them.

How large a photo can it handle?

Up to 25 MB and 50 megapixels. Neither is an upload limit, because there's no upload — they're what a browser tab can hold, since the file's bytes, the decoded image, the OCR engine's copy of the pixels and the finished .docx all have to exist in memory at once. A 48-megapixel phone photo fits. If you're over the line it's almost always a panorama or a scan at an unnecessary resolution; exporting at 4,000 pixels on the long side costs you nothing in accuracy.

Does the original photo go into the document?

Only if you leave that option on, and it's on by default because it makes proofreading easy: the text sits at the top, the photo underneath, and you can check a doubtful line against the original without switching windows. The picture is scaled to fit inside the page margins — never enlarged past its own resolution, which would only magnify the JPEG artefacts. Turn it off and you get a text-only document, which is much smaller and is what you want if you're pasting the result somewhere else.

Ratings & Reviews

Rate this tool

Loading reviews…

What is a JPG to Word converter?

A JPG to Word converter reads the printed text in a photo and gives you back a .docx with those words typed out as paragraphs you can edit. Photograph a letter, drop the photo on the page, and download a Word file you can change, quote from and send. This one does the reading in your browser tab. The photo stays on your device.

That definition sounds obvious, but the search results don't treat it that way. Two different jobs get sold under the same name. One wraps your picture in a .docx: the file opens, the photo sits in the middle of the page, and you can't edit a single word of it. The other runs optical character recognition (OCR). It works out which clusters of dark pixels are which letters and writes those letters into the document as real text. Only the second one is what you searched for.

Picture the person this page is built for. A landlord's letter arrived. You photographed it on the kitchen table because you want to quote two paragraphs back, correct one date and reply tonight. Retyping a full page on a phone is exactly what you're trying to avoid. To convert JPG to Word here, you drop that photo on the page and wait a few seconds.

How JPG to Word works

You drop, pick or paste an image. Five things happen next, all on your own machine:

The bytes are checked, not the name. The first 12 bytes of the file decide what it is. A JPEG starts FF D8 FF, a PNG starts 89 50 4E 47, and a WebP carries WEBP in its header. A PNG renamed to .jpg is read as the PNG it is. A GIF or a PDF gets told plainly that it isn't an image this page can open.
The image is opened to measure it. A phone photo comes out around 4,032 × 3,024 pixels. The size limits are checked right here, before the OCR engine is fetched, so an oversized panorama costs you no download.
Tesseract reads it. Tesseract is an open-source OCR engine, compiled to run in the browser. It returns the text plus a confidence score from 0 to 100.
The lines become paragraphs again. This is the step most converters skip, and the next section shows why it matters.
The .docx is written. Letter or A4, one-inch margins on every side, and the original photo underneath the text unless you switch that off.

The only thing fetched for the read is the engine itself. Tesseract's script, its WebAssembly core and the language model come from a public CDN the first time you use a language, and your browser caches them. The English model is about 5 MB. Your photo is never part of that traffic.

From OCR lines back to paragraphs: a worked example

OCR reports a page the way it sees it, one line of output per line of print. Pour that straight into Word and every printed line becomes its own paragraph. The document opens and looks right. Then you delete a word in the second line and nothing moves up, because a hard paragraph break sits at the end of every line on the page. Fixing that by hand across a full page takes longer than retyping it.

So the tool rebuilds the paragraphs before it writes the file. Here's a rent letter as OCR hands it over, nine lines of text, and where each line ends up:

Line as OCR read it	Characters	What happens to it
Dear Mr. Alvarez,	17	Paragraph 1 (a blank line follows)
Your rent will rise from $1,450 to $1,525 a month from	54	Starts paragraph 2; the longest line in its block
the first of March. The increase covers the build-	50	Joined; hyphen dropped, since the line runs to the margin and the next starts lowercase
ing insurance renewal and the new boiler.	41	Joined; a blank line ends paragraph 2
Please sign the enclosed form and return it by	46	Starts paragraph 3
February 14. A self-	20	Joined; hyphen kept, since the line stops well short of the margin
addressed envelope is included.	31	Ends paragraph 3
Kind regards,	13	Starts paragraph 4
Dana Whitfield	14	Joined onto "Kind regards,"

The result is 49 words in 4 paragraphs, and the page tells you so: "9 lines off the page became 4 paragraphs you can edit, with the photo underneath at 6.5 in × 4.88 in." "build-" and "ing" come back as "building". "self-" and "addressed" come back as "self-addressed". Two trailing hyphens, two opposite decisions, and both are right.

The rule in one breath: a blank line always ends a paragraph. A line that finishes a sentence and is shorter than 60% of the longest line in its block ends one too, because that's what the last line of a printed paragraph looks like. A trailing hyphen is dropped only when its line reaches that 60% mark and the next line starts lowercase. Everything else joins with a space.

Look at the last row, though. "Kind regards," ends in a comma, not a full stop, so the sign-off and the name join into one paragraph. It's one press of Enter in Word to split them. We'd rather tell you that here than have you find it in the file.

What you don't have to hand over

The image to Word converters that rank for this search share a familiar shape. You upload the photo to a server you know nothing about. You wait in a queue. Then comes the catch: a 5 MB cap unless you pay, a daily limit, an account before you can do more than one, or OCR itself parked behind a paid plan. Sometimes the Word file only arrives by email, which means handing over an address too.

None of that is a technical requirement. The recognition runs fine in a browser tab. The upload is there because the business model needs your file on its server, and the gates are there because a gate is how a subscription gets sold. For a photo of a lease, a payslip or a letter from your bank, that's a lot to hand a stranger in exchange for some typing.

This tool has no queue, no retention window and no server to keep your photo on. It takes files up to 25 MB, the same for everyone, and it works the same whether you're a student with a photographed textbook page, a freelancer with a client's printed brief, or someone at a large company who just needs one page retyped. That's the whole tool, not a sample of a bigger one.

When a JPG to Word conversion comes back wrong

On a flat, well-lit, straight-on photo of ordinary printed type, Tesseract gets close enough that you fix a handful of characters per page. When the result is worse than that, the cause is almost always the photo, and it's usually quick to fix. If your conversion didn't come out well, here's what to try first:

What you see	Likely cause	Try this
Whole lines of nonsense	Page shot at an angle	Shoot from directly above, square to the page
Letters merged together	Soft focus	Tap the text to focus, then shoot
Lines missing	Shadow or glare across the page	Move the lamp, or move the paper
"No text came back"	Handwriting, or a page far from upright	Straighten and relight it; handwriting won't read
Accents read as the wrong letters	Wrong language picked	Choose the page's language and let it read again
Two columns mixed together	Multi-column layout	Crop to one column per conversion

The tool also tells you when to be suspicious. If Tesseract's confidence comes back under 70, you get a note such as "OCR was about 69% sure of what it read, which is low — read the document through before you send it." The number is rounded down, so 69.6 reads as 69, never as a reassuring 70. If the long side of the image is under 1,000 pixels, you get a warning that the type is too small to read reliably. Around 300 DPI, roughly 2,500 pixels across a letter-size page, is where OCR gets dependable.

One quirk is worth knowing. A real compound that happens to break at the right-hand margin, like "well-" at the end of a full line with "known" below it, loses its hyphen and comes back as "wellknown". The rule can't tell that apart from a word the typesetter split. Skim any compound that sat at the edge of the page.

Paper size, the photo underneath, and the limits

The photo goes under the text by default, so a doubtful line and the original sit on the same page. It's scaled to fit inside the margins and never enlarged, since blowing up a photo only magnifies its blur. Here's where common images land:

Image	Pixels	Placed on Letter	Placed on A4
Phone photo, landscape	4,032 × 3,024	6.5 × 4.88 in	6.27 × 4.7 in
Phone photo, portrait	3,024 × 4,032	6.5 × 8.67 in	6.27 × 8.36 in
Letter page scanned at 300 DPI	2,550 × 3,300	6.5 × 8.41 in	6.27 × 8.11 in
Tall phone screenshot	1,170 × 2,532	4.16 × 9 in	4.48 × 9.69 in

The screenshot is the odd one out: it's limited by the page's height, not its width, so it comes out bigger on A4, the taller sheet. Switch the photo off and you get a text-only file, which is much smaller and is what you want if the words are headed somewhere else.

The limits come from what one browser tab can hold, not from a pricing page. The file, the decoded image, the OCR engine's copy of the pixels and the finished .docx all sit in memory at once. So: 25 MB per file, 50 megapixels, and 65,535 pixels on either side. A 48 MP phone photo fits. A 12,000 × 5,000 panorama doesn't, and the tool says so before it downloads anything: "That image is 60 megapixels, and a browser tab can decode about 50."

Related tools for photos, scans and PDFs

PNG screenshots work too: copy one and paste it anywhere on the page, and the image to Word read starts on its own. If your phone saved the photo as HEIC, some browsers can open it here and some can't. When yours can't, run it through HEIC to JPG first. WebP images work directly, and WebP to JPG is there if something else needs a JPEG.

For a scanned PDF, OCR PDF does the same reading across every page in one pass. If the PDF already has selectable text, skip OCR entirely: PDF to Word pulls the real characters out, and reading real characters always beats reading a picture of them. Extract Text from PDF does that when you only want plain text.

If you want the photo kept as a photo, JPG to PDF is the right tool. Once your edits are done, Word to PDF locks the layout before you send it.

Frequently asked questions

Why is the first conversion slower than the ones after it?

The first time you use a language, your browser downloads the OCR engine and that language's model. English, Spanish, French, Portuguese, Italian and Polish are about 5 MB each, German about 8 MB and Russian about 9 MB. After that they're cached, so the next photo in the same language skips the download and goes straight to reading. The read itself is real computation on your device, so a dense page takes a few seconds, a bit longer on a phone.

Why did two lines of my letter end up in the same paragraph?

Almost always because the first one doesn't end with sentence punctuation. A sign-off like "Kind regards," followed by a name, or an address block with no full stops, gets joined, since nothing on those lines says "paragraph ends here". Put the cursor after the comma in Word and press Enter. A blank line between the two lines on the printed page would have kept them apart.

If I switch to A4, does it read the photo again?

No. The tool keeps the scan in the tab, so changing Letter to A4 or turning the photo on or off rebuilds the Word file in a moment without running OCR again. Changing the language is different: it starts a fresh read, because a different model is reading the same pixels.

Can I convert JPG to Word for several photos at once?

Not in one go. Each conversion takes one image and gives you one .docx, named after it, so IMG_4471.JPG comes back as IMG_4471.docx. For a three-page letter, convert each page, then paste pages two and three into the first file. It's a few extra seconds, and each page gets its own confidence check on the way.

My photo looks sharp. Why am I warned it's too small?

Because sharp on a screen isn't the same as enough pixels per letter. An image that's 800 pixels on its long side looks fine as a thumbnail, but each character is only a few pixels tall, and Tesseract starts guessing. The warning appears under 1,000 pixels. Re-shoot closer, or export at full size. Upscaling a small image doesn't help, because it adds pixels without adding detail.

Is there an account, a limit per day, or a paid tier?

No account, no daily count, no tier. The reading happens on your device, so another conversion costs us nothing to allow. Microapp is paid for by members and by ads for everyone else, not by a gate in front of the thing you came for. And 10% of every dollar Microapp earns goes to charity, off the top, audited quarterly.