That’s true, though you can solve the image location problem by adjusting the LaTeX accordingly. I don’t actually use the images that Scrivener produces when compiling, so that aspect isn’t a problem for me. My setup for the user manual looks like this:
(all .sty files and such)
user_manual.tex -> (symlink to above and .tex folder inside)
All right, so the four cryptic folders at the top house my graphics. Mac-Black and White, Mac-Colour, Windows-Black and White, Windows-Colour. Depending on which user manual I’m compiling, I need different screenshots. Plus screenshots tend to change a lot, so using the embedded ones that Scrivener creates is way more trouble than it is worth—not to mention you can only have one, and not one out of four depending on compile mode, without a bunch of monkeying around with content filters and breaking sections apart wherever a screenshot appears.
The secret sauce is in the LaTeX preamble:
The all-caps in the middle comes from a custom MMD meta-data field, “PrintMode”, which I set to ‘mbw’, ‘mc’ or whatever according to the current version output. So in the actual output, I’ll get something like:
The key thing here is that this LaTeX command enforces this as a prefix to any included graphics in the document. So while Scrivener+MMD produces something like:
What actually gets used by the LaTeX typesetting engine is:
And suddenly my directory structure above makes sense. All I’m using from Scrivener is the name of the screenshot figure itself and the caption that it generates, the actual graphic being used is located one folder up and in one of the specialised print mode folders. All I need to do is change one small MMD meta-data field, and every single graphic in the compiled document suddenly switches from Mac to Windows (other things happen as well, such as different page layouts for the B&W versions, which aren’t used yet, but when we start selling hard copies of this thing, that will be what I use).
I can get away with this because whenever I add a screenshot, I create the four variations and put them into these folders and then drop a “placeholder” into the Scrivener project. I don’t ever rely on the placeholder to do anything except generate the appropriate syntax in the correct location of the .tex file.
All of this is probably a lot more complicated than what you need, but the same principle could be applied. You could set the graphics path to the compile folder so that even though you are typesetting from one folder up, they would be correctly included in the dynamic compile folder. All of that would give you a system whereby you can completely ignore the scrivener_generated.tex/ folder contents itself. All you would ever be interfacing with is the symlinked .tex file, and it would be interfacing with the compile folder via the graphicspath prefix, pointing to compile.tex/image_name.png.