Convert AZW3 to EPUB
Runs in your browser. No upload, no account.
Runs in your browser 0 bytes uploadedDrop your AZW3 file here
Nothing is uploaded. The book is unpacked and rewritten in this tab.
The text and the chapter order come across. The publisher’s fonts, styling and pictures do not — and the page below says why rather than pretending otherwise.
Inside an AZW3 is modern web markup, run through a mill
Amazon introduced KF8 — the thing in an .azw3 file — with the Kindle Fire in 2011,
and the interesting part is what it is made of. Where the format before it stored a book as one
long stream of markup from the 1990s, KF8 stores the real thing: proper XHTML, one document per
chapter, with CSS, embedded fonts, tables and drop caps. Structurally it is an EPUB. It was
built by taking one apart.
Then it was put back together the wrong way round. All those documents are concatenated into a single continuous stream, and the information about where each one begins and ends is moved out into two separate binary indexes — one listing the shell of each document, one listing the content that belongs inside it. Reading an AZW3 is therefore not a matter of decompressing text. It is a matter of reading two indexes and cutting the stream where they say to cut it.
AZW3 and EPUB, in brief
| Full name | Amazon Kindle Format 8 |
|---|---|
| Extension |
.azw3, .azw
|
| Format type | Amazon’s newer Kindle ebook — XHTML documents in a Palm database |
| MIME type |
application/vnd.amazon.ebook
|
| Full name | Electronic Publication |
|---|---|
| Extension |
.epub
|
| Format type | A reflowable ebook — a zip of XHTML with no fixed pages |
| MIME type |
application/epub+zip
|
An AZW3 to EPUB conversion keeps the chapter titles, and the indexes are why
This is the practical difference between the two Kindle formats, and it shows up in your library list. Because the documents survive as documents, the division a publisher set survives with them: chapter one is a file, chapter two is a file, and each carries its own title. Convert one here and the table of contents comes out reading Story of the Door and Search for Mr. Hyde, because that is what the file says. Converting the older MOBI format usually cannot do that — there are no documents to recover, only page breaks with nothing attached to them, so the chapters come out numbered.
Cutting a stream according to two indexes raises an obvious question: what if the pieces do not actually fit together? A misassembled book is the dangerous failure here, far worse than a failed conversion, because it reads almost normally — the sentences are fine, the paragraphs are fine, and a chapter break lands in the wrong sentence somewhere on page ninety. Nobody notices until they are lost. So the pieces are checked against the text they are supposed to cover, looking for gaps and overlaps, and if they do not tile it exactly the receipt says so rather than handing you something that looks finished.
Two other numbers get checked and reported for the same reason. The file states how many bytes of text it holds, and the unpacked result is compared against that figure. And the two indexes independently describe the same relationship — one says how many pieces each document has, the other says which document each piece belongs to — so they are cross-checked against each other. Agreement between two separate statements in the file is much stronger evidence than reading either one alone.
The one place where the stylesheets could have ended up in chapter fourteen
An AZW3's stream holds more than the book. The publisher's CSS is packed into the same run of
bytes, immediately after the last chapter, and a converter that simply reads to the end of the
stream drops several kilobytes of margin and text-indent into the
final chapter as though it were prose. There is a record in the file marking where the text
stops and the styling begins, and reading it is the difference between a clean last chapter and
a book that ends in stylesheet.
That styling is then deliberately not used, which is worth being straight about since the file went to the trouble of carrying it. A reflowable book is meant to be set by whatever you read it in — your type size, your margins, your line height, your choice of serif at two in the morning. A stylesheet converted out of a Kindle file specifies pixel margins and font stacks for a device you are probably not holding, and the result argues with your reading app instead of helping it. The words come across; the typography is yours.
A purchased AZW3 is more often locked, and how much more is unknown
AZW3 is the format the Kindle Store actually delivered in for most of the last decade, which
makes a file with that extension more likely than a .mobi to have arrived attached
to an account. More likely is a direction, not a proportion, and the proportion is not something
we can see: this page uploads nothing, so there is no record of what anybody hands it. A number
here would be a number we made up.
What can be measured is your particular file, immediately. A two-byte field near the start of every Kindle book says whether it is encrypted, so the answer arrives before anything has been unpacked — read out of a few hundred bytes rather than out of a few hundred thousand. If it says locked, this page stops there. The lock is tied to the account that bought the book, and this tool does not remove it under any framing: not as an option, not behind a confirmation, not with a pointer to somewhere that would. That is a different activity from moving a book between two containers, and only the second one happens here.
Plenty of .azw3 files are not locked at all. Standard Ebooks publishes in the
format. Publishers who sell DRM-free deliver it. Anything you produced yourself in Calibre is
unencrypted by construction — and if you are searching kindle to epub because a folder
full of books will not open on a new device, sorting them by which ones the header calls locked
is the fastest way to find out how much of the problem is actually solvable.
Extensions lie, so the header decides
Amazon's file extensions do not map cleanly onto the formats behind them. A plain
.azw is often the same KF8 layout as an .azw3, sometimes the older MOBI
one, and occasionally a container holding both so that old and new devices could each find
something they understood. On top of that, people rename files when a program refuses to open
one, so a folder of ebooks is partly a record of somebody troubleshooting.
Both Kindle formats begin with the same eight bytes. The only thing separating them is a version
number a little further into the header — six for the old design, eight for this one — and that
number is what decides how the file is read. The filename never enters into it. So an
.azw3 that is really the older format still converts, and the tool says which it
turned out to be and points at the page written for it.
Where Calibre wins, and where a tab wins
Calibre handles KF8 too, and handles more of it: the embedded fonts, the stylesheets, the illustrations, the library the book sits in. Twenty years of work, free, and the right tool if converting ebooks is something you do more than once. Its single requirement is somewhere to be installed, and that requirement is what this page exists around — an account without admin rights, a school-issued Chromebook, an iPad, a machine in a reading room, a laptop whose software is decided elsewhere. There the question is not which converter is best but which one can run at all.
Illustrations are the one real loss, and what it costs you depends on the book. A modern Standard Ebooks release carries a cover and a few decorative images; an illustrated children's edition carries dozens. Whatever your file holds is counted and reported before you download anything, precisely so the decision belongs to you instead of arriving as a surprise. After that the EPUB is an ordinary EPUB: set it on paper at a size you choose, or go the opposite direction with something that began life as a document. None of this leaves the tab, and your browser will confirm it — the transfer counter for this page does not move while a five-megabyte book is taken apart in front of you.
Questions
My .azw3 came from Amazon. Will it convert?
Only if nobody locked it, and the tool tells you which within a millisecond of receiving the file — the answer sits in a field in the first few hundred bytes, so nothing has to be unpacked to read it. AZW3 is the format the Kindle Store delivered in for years, so a file that arrived that way is more often locked than a .mobi you found in a library of free books. More often is not always: publishers who sell DRM-free deliver .azw3 too, and so does anything you built yourself. This page does not remove a lock under any circumstances, so the check is the whole of the answer.
Why do the chapters keep their names here?
Because AZW3 still contains them. Amazon’s newer format packs a book as separate XHTML documents — one per chapter, with its own title — and then runs them together into a single stream with two indexes recording where each begins. Reading those indexes gives back exactly the division the publisher set, titles and order included. The older MOBI format has nothing comparable: it is one continuous document whose only structural marker is a page break, which is why converting one often produces numbered sections rather than named chapters.
How do I know the whole book came across?
Two checks run every time, and both are reported rather than assumed. The file states how many bytes of text it holds, and the unpacked text is compared against that number — a decompression that stopped early cannot pass silently. Then the document boundaries from the two indexes are checked for gaps and overlaps against the text they are supposed to cover; if they do not fit together, the receipt says so instead of handing you a book whose chapters are quietly stitched in the wrong places. That second failure is the dangerous one, because a misassembled book reads almost normally.
What about .azw files, without the 3?
Drop one in and see. The extensions Amazon has used do not map cleanly onto the formats inside: a plain .azw is often the same KF8 layout as an .azw3, sometimes the older MOBI one, and occasionally a container holding both. Since the format is read out of the header rather than the filename, whichever it turns out to be gets handled correctly, and the tool names what it found. The same is true in the other direction — a file someone renamed to .mobi because a program refused to open it still converts here.
Where did the pictures and the fonts go?
They stayed in the original file, which is worth knowing before you throw it away. Getting an image out of the archive is easy; getting it into the finished book is not, because every picture needs its own declaration in the EPUB’s manifest and its own reference from the chapter that used it — a job of its own rather than a line of code. So the count is reported to you and nothing vanishes unannounced. The fonts and stylesheets are a different case: those are omitted on purpose, because the whole point of a reflowing book is that your reading app decides the typography, and a stylesheet built for a Kindle screen will argue with it.
Is this the same thing as removing DRM?
No, and the difference is not a technicality. Removing DRM means defeating encryption to get at content someone locked; this converts a book that was never locked from one container into another. If a file is locked, this page stops and says so — it does not offer a way around it, ask you to try a key, or bury the option behind a warning. The two operations are often discussed together because the same files are involved, and they are not the same operation, and only one of them happens here.
Does the file get uploaded anywhere?
No. The header is read, the text records are unpacked, the two indexes are consulted and the EPUB is written by code executing inside this page. That is checkable rather than something to take on trust, and the sharpest check here is a counter rather than a list: developer tools report total bytes transferred for a page, and handing over a five-megabyte book moves that figure by nothing at all. One fetch happens on your first conversion, which is this page collecting the code that does the work. After that there is no traffic in either direction, which is why a second book converts with the machine unplugged.