[TOOL] PureEPUB: Free web tool to convert messy PDFs & EPUBs into native Kobo KePub

[lones]

Thanks again everyone, this round of feedback was really useful. Today's update is live, here's what changed:

Originally Posted by JSWolf

Remove the overall L/R margins. Make them 0. Also, remove line-height.

Done, as promised. The Kobo/KePub (and generic EPUB) stylesheet no longer sets a body line-height or margins, so your device settings are in full control now. The Kindle output is unchanged.

Originally Posted by fogice

I found that, but it didn't work well (I don't remember why, maybe it crashed partway through).

You were right, and I found the cause. With "Keep Images" on, a scanned book could take longer than the server's time limit, so the conversion got cut off midway. On the 1927 Internet Archive PDF graycyn tested, a full conversion with Keep Images on used to run past 3 minutes and get cut off; on the live site it now finishes in about 20 seconds, upload included. Would love it if you could give it another try.

Originally Posted by fogice

Also, I'd like the website to keep my defaults. I never want my covers cropped or stretched.

The site now remembers your settings in your browser: target device, cover screen size, cover layout, chapter splitting and Keep Images. Pick "Original Ratio" once and your covers stay uncropped on every visit.

Originally Posted by graycyn

Many sentences were broken apart, this seemed to happen a lot during dialog passages.

Thank you for the detailed test, Tawny turned out to be a great test book. I've added it to my test set. Fixed in this update:
- A line starting with an em-dash and a lowercase word ("…the dogs who ran / — never house dogs…") no longer starts a new paragraph. That was most of the broken dialog sentences in Tawny.
- Two more cases that split sentences in other books: a quoted word landing at the start of a line, and narrow-column books being mistaken for poetry.

Originally Posted by graycyn

One typographic annoyance was spaces on either side of every em-dash.

In this PDF the spaces are actually in the Internet Archive text layer itself, but for English books the converter now closes them up ("days—not"). A word glued to an em-dash at the end of a line also stays glued now.

Originally Posted by graycyn

the apostrophes were NOT curly.

Fixed, apostrophes inside words are curly now (can’t, dog’s), matching the quotes.

Originally Posted by graycyn

I did check, and italics were not preserved.

Not yet, I'm sorry. Italics are on my list. Carrying them over from the PDF needs a bigger change to how text is extracted, so it's a separate piece of work.

Originally Posted by JSWolf

I would not use this for PDF.

Fair enough, PDF is the hard case and it will never be perfect. Born-digital PDFs come out quite clean though, and it's improving every week thanks to tests like the ones in this thread.

Thanks again, keep the bug reports coming!

[JSWolf]

Sounds good. I'll give it another go when I get a chance. Thanks.

[fogice]

The new version works a LOT faster, but it ignores all embedded graphics in a PDF. I turned the beta photo option on in the settings.

[lones]

Originally Posted by fogice

The new version works a LOT faster, but it ignores all embedded graphics in a PDF. I turned the beta photo option on in the settings.

Thanks for the report!

My guess is that's exactly what's happening here: the current photo option only grabs embedded raster images (JPEG/PNG, photos, scans). If those graphics or charts were drawn as vector shapes (very common in PDFs exported from Word or LaTeX), the extractor just doesn't see them as standalone images yet.

I'm actually planning to build a renderer pass for this—rasterizing those vector graphics and page diagrams on the fly so they can be injected straight into the EPUB like regular images.

Still, if you don't mind sharing the PDF (or just a couple of pages with those graphics), it would be super helpful to test the new logic against!