iloveepubPrzetwarzane lokalnie — pliki nigdy nie opuszczają Twojego urządzenia
Kompresja EPUB
Zmniejszaj pliki EPUB, przekodowując obrazy i usuwając zbędny balast — w całości w Twojej przeglądarce. Za darmo, bez wysyłania, bez limitów.
01Wybierz EPUB
When a smaller EPUB is the actual job
Amazon documents two different Send to Kindle caps. Email delivery rejects an attachment over 50 MB. The Send to Kindle website accepts a file up to 200 MB, while the desktop and mobile apps still stop at 50 MB. A scanned cookbook or a manga volume often lands between those two numbers, so the same book the website would take still bounces from a Kindle email address. The E-reader preset is aimed at that gap: JPEG figures are recompressed, and anything whose long edge is over 1400 px is scaled down, which is the neighborhood of a 6–8 inch 300 ppi e-ink panel rather than a print-resolution scan.
The same pass is what you want for a library loan that has to fit on a Kobo with little free space, or for an InDesign proof whose chapter figures are several thousand pixels wide while the cover is already fine. If the book is mostly text, or its images are already small, the analysis says so before you download anything. Compression cannot invent savings that are not in the pictures.
What the browser rewrites
The archive is unzipped in memory. Chapter XHTML, stylesheets, fonts, and the OPF metadata are copied through. Only JPEG and PNG images are candidates for change. A JPEG is re-encoded on a canvas. A PNG is re-encoded only when it is also scaled down, and it stays a PNG so the manifest href does not have to change. GIF, WebP, AVIF, and BMP are left byte for byte. The cover is treated more gently than interior figures: it is not downscaled, and its JPEG quality is raised by 0.1 and capped at 0.9, because shelf thumbnails are drawn from that one file.
The book is then packed again. The mimetype entry is written first and stored uncompressed, which is the check strict readers actually perform. Already-compressed media and fonts are stored rather than deflated a second time. macOS junk (__MACOSX, .DS_Store) and Windows junk (Thumbs.db, desktop.ini) are dropped. If the repacked book is not smaller than the file you opened, the download is the original. You do not get a larger file with the word compressed in its name. Lossless only strips that junk and repacks. Balanced uses JPEG quality 0.8 and a 1920 px long edge. E-reader uses 0.76 and 1400 px. Extreme uses 0.6 and 800 px. Target size retries from the original images down that ladder and stops when the file fits, or tells you it could not.
What compression will not do
Embedded fonts are the real bulk of many Word and InDesign exports, and this page does not subset or remove them. Audio, video, and WebP stay as they are. DRM-protected files are refused. The cap is 120 MB, because the whole archive has to live in the tab. A larger book needs a desktop tool.
Target size is best effort, not a promise. Each round starts from the original images, so JPEG loss is not stacked, but if those images are already efficient the result can stay above the number you typed. The page says the target was missed instead of pretending the file fits.
Calibre does a different job
Calibre's Convert books and Polish books can subset fonts, downsample with another encoder, rewrite CSS, and send the result to a device. That is the right next step when the weight is in fonts, or when you also need AZW3. It is also a full conversion pipeline, and it can change typography you did not mean to change.
This page only recompresses images and repacks the zip. Text nodes are not rewritten, so links, footnotes, and the publisher's hyphenation stay put. There is no device driver. If Lossless already fails to shrink the file, a stronger JPEG preset will not help. Font tools in Calibre are the honest next step.
What each preset changes
| Preset | Images | Long edge |
|---|---|---|
| Lossless | Not re-encoded. Junk files dropped, archive repacked. | Unchanged |
| Balanced | JPEG quality 0.8. PNG re-encoded only when scaled. | 1920 px |
| E-reader | JPEG quality 0.76. The cover is not scaled. | 1400 px |
| Extreme | JPEG quality 0.6. Visible softness in figures. | 800 px |
| Target size | Retries from the original images, stronger each round. | Down to 800 px |
Jak skompresować EPUB
Wybierz plik EPUB
Otwórz plik ze swojego urządzenia. Przeglądarka czyta go bezpośrednio — nic nie jest nigdzie wysyłane.
Wybierz profil kompresji
Wybierz czyszczenie bezstratne, profil pod e-czytniki albo docelowy rozmiar pliku w MB. Dla każdej opcji szacujemy wynik, zanim się zdecydujesz.
Pobierz mniejszy EPUB
Skompresowana książka zapisuje się bezpośrednio na Twoim urządzeniu. Oryginalny plik nie jest nigdy modyfikowany.
Częste pytania
Why does Send to Kindle email reject a file the Kindle app would store?
Amazon documents a 50 MB cap on Send to Kindle email attachments, and a 200 MB cap on the Send to Kindle website; the desktop and mobile apps share the 50 MB email cap. A book between those sizes has to shrink before it can go through the email address, even though the same Kindle could hold it over USB or the website.
Which preset matches a 6-inch Kindle or Kobo?
E-reader. It caps the long edge at 1400 px and uses JPEG quality 0.76. A 300 ppi panel in the 6–8 inch range does not benefit from print-resolution scans. The cover image is not scaled down with the interior figures.
Do chapter text and the table of contents change?
No. XHTML, CSS, navigation documents, and metadata are copied. The bytes that can differ are JPEG and PNG images, dropped junk files, and the zip packaging, including a mimetype entry stored uncompressed in first position.
The file came back the same size. Did it fail?
Not if the images were already small, or the book is text and fonts. The download stays the original whenever a repack would not be smaller. Fonts are not subsetted here. Calibre can do that on a desktop, and this page will not pretend a JPEG pass can.
Can I force an exact megabyte target?
Target size retries from the original images until the archive is under your number or the ladder runs out. It does not stack generations of JPEG loss. If it cannot reach the target you still get the smallest of those attempts, and the page says the target was missed. Files over 120 MB are refused before that work starts.