[Jaws]

I'm uncertain whether this is truly a "from KFX" thing or something broader in Calibre; it seems a lot more common in conversions from KFX, but that may just be because of the way some publishers are doing their own conversions in-house for older books.

The problem:

After conversion, illustrative graphics are often in separate files (with separate generated styles, too) from the accompanying text. This often ends up splitting logical chapters (often relatively small ones, like individual recipes in illustrated cookbooks) into three or more separate files. This does not play nice with some readers, and is particularly annoying when the graphic is actually quite small.

The resulting structure might look like this (artificial names for clarity):
file001.xhtml (the chapter number and title)
file002.xhtml (a graphic that on a printed page is right with the chapter number and title, ranging from a 50x50 icon up to a 3/4-page illustration)
file003.xhtml (the text of the chapter, sometimes itself broken up similarly if there are interior illustrations even interspersed with running narrative text and no subheadings)

And on many readers (such as at least ten Android and Windows based readers that I've examined, and futilely attempted to customize), each file is displayed as if it contains a CSS "page-break-before," which is both really annoying and separates sometimes-meaningful illustrations from the accompanying text. (Cookbooks and biological-sciences texts are especially annoying about this; one conversion from a 1980s-era original print publication broke a 17-page chapter into 83 separate files, making it almost impossible to read.)

The question:

There are probably conversion parameters somewhere — perhaps in Calibre as a whole, more probably in KFXInput or even the KFX specification — that are creating this kind of chaos. Are these user-adjustable (and where and how), or is this something I'm just going to have to manually edit the internal structure by merging files in the editor after conversion?

I'm not necessarily suggesting changes to either Calibre or KFXInput, just looking for ways to automatically import sources that aren't straight running text without spending hours rebuilding single chapters so that they'll display illustrations in context. Admittedly, given the way publishers are munging their conversions, there is going to be at least some of the latter no matter what.