Image missing after conversion from docx to epub

[kefren]

I was converting from docx to epub. But for some reason one of the images (the publisher colophon) is disappearing after conversion.

I use the standard epub options, apart from adding the cover and metadata, and ticking epub “Preserve cover aspect ratio”.

In the Word docx it is a png file (I was told they're best for ebooks - but I get the same issue if I switch to jpg).

Strangely - another of my files has a number of png images, and Calibre converts them all into the epub with no problems, apart from this one image at the start of the book, which it ignores. There are no error messages.

I found a workaround. First I tried saving the docx as a web page; then I zipped up the html file and folder of images Word created as a single file to drag into Calibre. I converted to epub - image still missing in the epub (even though I can see it in the zip file subfolder).

Then I tried again, this time repeating it but saving it from Word as "Web page - filtered". And Calibre is able to convert that and display the png colophon with no problem! Which raises various questions:

1. Any idea why this works? As in, what does Calibre see differently from the docx and basic html versions? Is it a bug in Calibre that I need to report?

2. I could add this to my standard procedure, but could the extra stage of saving the doc as a "Web page - filtered" first cause other problems (e.g. in layout) while fixing the missing colophon? I'm always wary of adding extra stages, since it always seems like an extra chance for weirdness.

I switched to Calibre as a way round having to rely on the Smashwords/Pronoun/KDP converters, and thought I'd found a nice simple system in using it to generate my own docx>epub ... until this!

If I have overlooked anything obvious as to why that image is being ignroed by Calibre I'd be grateful. This task has taken days so far, I had expected it to only take an afternoon. (Bad mistake, I know.)

Oh, all this has been replicated by someone else using my source files. I've attached the files I can, but strangely even though Calibre will convert from docx (not doc), this forum won't let me upload docx, only doc files, so it won't let me attach one of the files I converted from as an example ...

Finally - I was on an old version of Calibre from 2013, but uninstalled it and installed the 64 bit 2.67 version today. Same problem.

[kefren]

Worked out a way to attach the docx source - in a zip file (attached!) Also the png file I was using.

I have cut the document back to just the first page with the image - it still has the same problem when converted to epub.

[BetterRed]

@kefren - that book is under copyright. The owner of Mobileread does not allow posting of such books - suggest you remove it before someone else does. Even if you own the IPR I suggest you remove it.

If you raise a ticket at Bugs : calibre you can attach the DOCX and EPUB and mark the ticket as private.

FWIW - as soon as Kovid produced the DOCX Input facility I said good riddance to Word's Filtered HTML.

PS : welcome to Mobileread

BR

[kefren]

I'm the copyright owner and author (Karl Drinkwater), so can post/share/sell it. I had just updated the contents with this extra image so I can upload the new version to my distributors.

However, just so this doesn't cause problems, I've removed the full docx file and just attached a page with my name and the title and image - the same behaviour shows when I convert it to epub, so the behaviour can still be replicated but with no issues about whether it is really my book! I only attached the full docx originally because I thought it would help, and because I have the rights to do that.

As to the FWIW: I'd hoped docx input had solved my problems too, until this pesky image! Thanks.

[kefren]

Bah, I spoke too soon - the epubs created by my workaround (filtered html, zipped up and imported into Calibre to be converted to epub) then fail validation at http://validator.idpf.org ... 　
Back to square one.
Looks like my only option is to go from docs>epub, and accept that Calibre is ignoring some of the images. This has been a very frustrating few days!

[BetterRed]

@kefren - I intuited you were the author, that's why I didn't remove the attachment toot sweet

Long shot 1 - import the DOCX into the calibre book Editor
Long shot 2 - install the e-Book Tools - a Word add-in - it can create epubs - and a whole lot more than that.

BR

[kefren]

1 - When I dragged a docx in and clicked on "Edit book" in Calibre it told me it could only edit AZW3 or epubs. So I converted to epub, then clicked "Edit book", but the strange thing is, in the images section there is only the cover, not the publisher colophon. So Calibre's conversion to epub removes it totally.

2 - Will have a look, thanks!

[BetterRed]

you need an empty book - start the editor from where its installed eg program files\calibre2\ebook_edit.exe, then File->Import HTML....



BR

[kefren]

Originally Posted by BetterRed

you need an empty book - start the editor from where its installed eg program files\calibre2\ebook_edit.exe, then File->Import HTML....

Attachment 151645

BR

Ah. I just did that - same issue. Even though the image shows up in the list of files on the left, it is blank and doesn't show up in the preview on the right. (Image attached).

I also tried the Word plugin - after fiddling with the settings I got an epub with the correct styles, but there were other issues with is instead (broken navigation and so on), so I think I'm likely to end up in an even bigger tangle with that.

The clean Calibre docx>epub dream had been so good until it ran into issues with this image!

[BetterRed]

I'm out of ideas - hopefully someone will logon soon with good knowledge of the inner workings of DOCX and EPUB formats

BR

[kefren]

Thanks BetterRed. I'm currently playing around with it. Strangely, switching the logo in my docs for a different png file sometimes works for Calibre. So maybe it is possible that Calibre doesn't like my png file and the jpg based on it, though I have no idea what would be different about it. Seems to be replicable. I'll hope someone looks at my files and works out what Calibre is "thinking"!

[jackie_w]

Looking at the calibre Job details for the DOCX -> EPUB conversion gives the following clue

Detected an image that looks like a cover

I'm far from expert in calibre code but a quick look at some of the DOCX cleanup code shows that calibre does some basic tests to try to decide whether the first image in the source document may be a cover, including:

• checking the HTML for the page where the image appears to see how many 'other' tags precede the <img src="first_image" /> tag, i.e. is the image very near the top of the page.
• is image height/width ratio between 0.8 - 1.8
• is image height x width > 160000, i.e is it a large image

It looks like your logo image may be (at least???) 1500x1767 which is pretty big for a logo. I suggest adding your logo as a much smaller image and trying the conversion again. I don't have a new enough version of MSWord to try it myself.

In the long run you'd need to discuss the situation with Kovid. I don't know how the first-image-might-be-a-cover rules were originally decided and whether they could be changed.

[jackie_w]

Follow-up...

Whilst the above may be true the way to fix your problem is much more simple. In the calibre conversion settings:

DOCX Input option: check the box labeled 'Do not try to autodetect a cover from images in the document'

[theducks]

If that is the only major problem, why FIGHT getting DOCX (or any other) to work perfect.

Just place the image in the editor, The same type of things you normally would use the Editor for... Touch-up

[kefren]

Jackie_W you are a genius! Yes, ticking that setting on docx input fixes it; on top of fixing the problem, you also worked out what the cause was (which makes sense in hindsight - for example, if I didn't bother adding a cover in one of the tests, Calibre chose the logo as a cover, and if it wasn't there it created a random design with the title on - probably because of the behaviour you spotted). It also explains why some of my tests worked e.g. with other images - they were probably just lower resolution, rather than it being anything to do with a dodgy png.

This is great. I'll lower the resolution of the logo (a hangover from the 300dpi requirements for the print copy) and also add to my notes to tick that option.

In a way it is funny that over the last few days I have done about a hundred conversions, installed several new bits of software (none of which worked), and edited stacks of trial images - and the answer all along was a tick box! I generally avoid ticking things I don't understand, probably with good reason, and keep settings as close to the default as possible, but that's one I should have known about.

I can't thank you enough. If you want (correctly-formatted) epubs of any of my books you are welcome! http://www.karldrinkwater.uk/