PNG to Word

Which mode do you actually want?

"Convert PNG to Word" means two different things depending on the image, and picking the wrong one is the usual reason people are disappointed by the result.

Editable text runs OCR — optical character recognition — over the picture, identifies the letters and numbers, and writes them into the document as real Word paragraphs. Afterwards you can edit, restyle and spellcheck them like anything you typed. Use it when the words are the point.

Picture in document skips OCR entirely and places the image into the .docx, scaled to fit the page width. The text inside it stays part of the picture, so it is not editable — but the appearance is preserved exactly. Use it for charts, diagrams, logos, and screenshots you want to show rather than quote.

What structure survives, and how it is decided

Most free image-to-Word tools hand back one paragraph per printed line. That is not a document — every wrapped line is its own paragraph, so the text will not reflow when you edit it, and the headings are just larger sentences. This page infers the structure instead, and the rules are worth knowing because they explain the failures as well as the successes.

Paragraphs

Two consecutive lines belong to the same paragraph when the first one ran out to the right margin, or when it did not finish its sentence. Neither test alone is enough: the lines of one paragraph ended between three and six character widths apart in our measurements, which no margin tolerance separates cleanly, while a line ending "…Holdings Ltd." has a full stop that ends nothing. Together they are reliable.

On a page whose text occupies less than half the image width — a receipt, a label, a menu, a form — nothing is joined at all, and every printed line stays its own paragraph. There is no evidence of wrapping on such a page, so joining lines would be guessing.

Headings

The body size is whichever type size covers the most characters on the page, not the most lines — so a page of small print under one large title still resolves to the small size. A line becomes a heading only if it is larger than the body and is a single short line. A standfirst set large and wrapped over three lines is body text set big, not a title; a rule that looked only at size would promote its last line to a subheading. The heading sizes present on the page are then ranked, so the largest becomes Heading 1 and the next Heading 2.

Lists

A bullet is a token that repeats at a shared left edge. Which character the engine reports cannot be predicted — in our fixture a printed • came back as "+" on two lines and "*" on a third — so the character is not the signal, the repetition is. Indentation is deliberately not used: a first-line paragraph indent is the commonest indent in typesetting, and keying off it would turn the first line of every paragraph into a bullet. The cost is that a list with a single item comes back as an ordinary paragraph.

All of this is measured rather than asserted. Eight pages with a known structure are rendered, read with the same engine this page runs and scored: 30 blocks out of 30 correct, with each rule confirmed to break at least one of them when it is removed. Prefer one paragraph per printed line? Choose Plain paragraphs.

Capabilities at a glance

What this page does and does not do, stated plainly. The "No" rows are the point, not an omission.

Capability	Supported	Detail
Input
PNG, JPG, WebP, BMP, GIF, TIFF	Yes	Any raster image the browser can decode.
Scanned PDF	Yes	Each page is rendered and read; a page range narrows it.
Several files at once	Yes	One .docx per file, or one combined document.
HEIC from an iPhone	No	No browser decodes HEIC; convert it first with our HEIC to JPG tool.
File size limit	No	Bounded by your device’s memory, not by a plan.
Recognition
Languages	Yes	35 single languages plus 4 two-language combinations.
Headings	Yes	Inferred from type size and line length; mapped to Word Heading 1–3.
Bullet lists	Yes	Detected from a bullet repeated at a shared left edge.
Paragraph reflow	Yes	Wrapped lines are rejoined so the text reflows when you edit it.
Numbered lists	No	A leading digit is indistinguishable from a figure in a table.
Tables	No	Use our PDF to Excel tool, which reads columns properly.
Handwriting	No	The engine is trained on printed type.
Word output
Real .docx paragraphs	Yes	Built with the docx library, never a .txt renamed .docx.
Editable and searchable text	Yes	Select, restyle and spellcheck it like anything you typed.
Picture placed in the page	Yes	A second mode that skips OCR and keeps the appearance exactly.
Original fonts and colours	No	Word’s default body style is used; only the structure is carried over.
Columns and page geometry	No	A two-column page comes back as one column in reading order.
Embedded pictures inside recognised text	No	Text mode writes text only; use Picture mode for the image.
How it runs
Nothing is uploaded	Yes	The image and the document never leave your device.
Signup	No	There is no account and no daily limit.
Watermark	No	Nothing is added to the document.
Offline use	No	The engine is fetched from a CDN on first use, then cached by the browser.
API	No	This is a page, not a service.

Scanned PDFs and photographs

A scanned PDF is a picture of a page inside a PDF wrapper, so it needs exactly the same treatment as an image. Add it here and each page is rendered at high resolution and read; a range like 1-3, 7 limits it to the pages you want. If you can select the text in a PDF reader the document already has a text layer, and PDF to Word will copy it across exactly instead of recognising it from pixels — always the better result when it is available.

Photographs of paper are the hardest input. Blur, glare, shadow and skew cost far more accuracy than small type does. Shoot square-on in even light, fill the frame with the page, and the result improves more than any setting on this page can manage.

Languages

Thirty-five languages across Latin, Cyrillic, Greek, Hebrew, Arabic, CJK and Indic scripts, plus four two-language combinations for pages that mix English with another language. Pick the language before converting: running English over a German page costs accuracy on every accented word. Each language is a separate download the first time you use it, which is why the list is a choice rather than automatic detection.

What this page will not do

Tables. Recovering a table means deciding which number belongs in which column, and a tool that guesses wrong produces a document that looks right and says something false. PDF to Excel does that job properly.
Numbered lists. A leading "1." cannot be told apart from a figure at the start of a line, so numbers are left in the text where they were.
Columns and page geometry. A two-column page comes back as one column in reading order.
The original fonts and colours. The document uses Word’s default body style; only the structure is carried across.
Pictures inside the recognised text. Text mode writes text only.
Handwriting. The engine is trained on printed type.

Where your image goes

Nowhere. There is no upload step: the OCR engine and the Word builder both run in the browser, on your device, and the page contains no code that could send a file anywhere. You can check that yourself in the Network tab — the only requests are for the engine and its language data, from a CDN, and they are cached after the first use. Our analytics and ad scripts load like they do on any page and see the page view, never the file.

Frequently asked questions

Yes, completely — no signup, no watermark, no page limit and no daily cap. The conversion runs in your browser rather than on a server, so there is nothing to meter. Most competing tools allow one free conversion a day before asking for an account.

In Editable text mode, yes — OCR writes the words into the document as real paragraphs you can select, edit, restyle and spellcheck. In Picture in document mode the image is placed as a picture, so the text inside it stays part of the image and is not editable.

Yes, and this is the part most free converters skip. A heading is recognised from its type size and the fact that it is a single short line, then mapped to Word Heading 1, 2 or 3. Wrapped lines are rejoined into one paragraph so the text reflows when you edit it. A bullet that repeats down the page at the same left edge becomes a real Word list. Turn all of it off with Plain paragraphs if you would rather have one paragraph per printed line.

By type size and line length together, never by size alone. The body size is whichever size covers the most characters on the page, so a page of small print under one large title still resolves correctly. A line only becomes a heading if it is larger than the body AND is a single short line — a whole paragraph set in a large face is body text set big, not a title, and a rule that looked only at size would turn its last line into a subheading.

A list is recognised from a bullet character that repeats at a shared left edge. A list with a single item has nothing to repeat, so it is left as a paragraph — deliberately, because one line starting with a dash or an ampersand is far more often ordinary prose. Numbered lists are not detected at all: a leading "1." cannot be told apart from a figure at the start of a line.

No. Both the OCR engine and the Word builder run in your browser, so the image never leaves your device. That makes this safe for contracts, payslips, IDs and private screenshots. The only network requests the page makes are for the engine itself and the language data, which come from a CDN and are then cached.

Yes. Add the PDF instead of an image and each page is rendered at high resolution and read. A page range like 1-3, 7 limits it to the pages you need. If the PDF already has a text layer — if you can select the text in a PDF reader — use our PDF to Word tool instead: it copies the text across exactly rather than recognising it from pixels.

Yes. Drop as many as you like. Choose one document containing all of them, with each file under its own heading, or a separate .docx per file downloaded as you go. A file that cannot be read is named in the list rather than silently dropped.

Screenshots and flatbed scans are usually near-perfect — measured at 100% character accuracy on our own English, German and Arabic fixtures. Photos of paper are less reliable: blur, glare, shadow and skew cost far more accuracy than small type does. Check the recognised text in the box before you download; it is editable for exactly that reason.

Thirty-five, in Latin, Cyrillic, Greek, Hebrew, Arabic, CJK and Indic scripts, plus four two-language combinations for pages that mix English with another language. Pick the language before converting — running English over a German page costs accuracy on every accented word.

No, and no browser tool honestly can. You get the words in reading order with their structure, not a rebuilt page: a two-column page comes back as one column, tables lose their columns, and the document uses Word’s default body style rather than the original typeface. If the appearance matters more than the words, use Picture in document mode, which keeps it exactly.

Because doing it badly is worse than not doing it. Recovering a table means deciding which number belongs in which column, and a tool that guesses wrong produces a document that looks right and says something false. Our PDF to Excel tool does that job properly, reading columns from the whitespace between them or from the ruling lines when the table is ruled.

Choose Editable text when the image is a screenshot of a document, an article, a letter or a receipt and the words are the point. Choose Picture in document when the image is a chart, a logo, a diagram or a photo, or when the exact appearance matters more than the text.

In Editable text mode the browser downloads the OCR engine and the language data the first time, then caches them, so later conversions start quickly. Picture in document mode has no OCR step and is near-instant.

Yes. Any raster image the browser can decode works: PNG, JPG, WebP, BMP, GIF and TIFF. An iPhone HEIC does not — no browser has a HEIC decoder — so convert it with our HEIC to JPG tool first.

Yes. It is a real Office Open XML document built with the docx library, not a text file renamed .docx, so Word, Google Docs, LibreOffice, Pages and every mobile viewer open it. Headings come through as real heading styles, so the document outline works.