I think that if this were tried, and the results examined, you would see why that would not be an optimal way of working. Here is what some sample .tex looks like, using a best-practices approach. First, in the preamble somewhere we set up the break macro so that it prints an image, horizontally centred within the text block.
\newcommand{\sectionbreak}{
{\centering \includegraphics[width=232pt]{scene_separator.png}}
}
You’d have this somewhere at the very top of the document in your preamble. Then each section break throughout the book would look like this:
\part{Red Book}
\label{redbook}
Dri srung gronk ozlint; zeuhl la, ti dri. Relnag xi nalista dri lydran wynlarce, prinquis zorl nalista, zeuhl re obrikt relnag erk wynlarce wex pank gronk? Menardis clum, morvit xu ma yem twock irpsa ma cree tolaspa. Erk teng flim obrikt; menardis nix frimba tharn nalista kurnap rhull.
\sectionbreak
Relnag sernag arul; zorl ma ux flim ma wex yiphras prinquis, ti obrikt twock. Irpsa groum xu frimba re irpsa harle quolt velar ewayf teng, korsa tharn srung re lamax. Srung nalista; groum ma erc, erk la gra prinquis vusp fli vusp frimba lydran vo urfa teng berot ju quolt. Athran xi, gen velar srung; lydran su zeuhl fli jince su xu fli. Ma vo whik arka teng nix ozlint vusp pank.
It’s clear what is meant, it is easy to work with directly, and even write with, and nobody is going to question why you did things this way.
If instead you were to use Scrivener to insert an image between every single break, using its Separators pane to inject an image code with some additional formatting around it to present it properly, every single section break in the book would look like this:
\part{Red Book}
\label{redbook}
Dri srung gronk ozlint; zeuhl la, ti dri. Relnag xi nalista dri lydran wynlarce, prinquis zorl nalista, zeuhl re obrikt relnag erk wynlarce wex pank gronk? Menardis clum, morvit xu ma yem twock irpsa ma cree tolaspa. Erk teng flim obrikt; menardis nix frimba tharn nalista kurnap rhull.
{\centering \includegraphics[width=232pt]{scene_separator.png}}
Relnag sernag arul; zorl ma ux flim ma wex yiphras prinquis, ti obrikt twock. Irpsa groum xu frimba re irpsa harle quolt velar ewayf teng, korsa tharn srung re lamax. Srung nalista; groum ma erc, erk la gra prinquis vusp fli vusp frimba lydran vo urfa teng berot ju quolt. Athran xi, gen velar srung; lydran su zeuhl fli jince su xu fli. Ma vo whik arka teng nix ozlint vusp pank.
It’s messy, and bad practice to “repeat yourself” over and over like that. We have the ability to make commands to avoid that pitfall, and to be able to make central changes to how those commands work, if need be.
Beyond that, I would say there is a philosophical point to be made, at least as a matter of personal preference. That is not the type of detailing I would use Scrivener for, personally, and rather see it entirely a design decision to be solved with the .tex setup. The image itself would probably not be in Scrivener at all, save maybe as an aliased binder link to the texmf folder copy. For example the L&L logo on the copyright page of the user manual is found nowhere in my Scrivener project (nor indeed is anything on that copyright page, except a few pieces of metadata driving revision text and such).
Don’t get me wrong, you could do things the way you describe, using Scrivener’s <$img:...> code in the separators pane—it just wouldn’t produce a clean .tex file that conforms to best practices, and as I say, more ideologically speaking, dips into depending too heavily upon the wrong tool (Scrivener) to design the document. At least that’s how I feel about it. If you’re going to use LaTeX, you might as well use LaTeX, and not some chunky hybrid. 
I’m pretty sure that is what Scrivomatic uses as well, so you might want to check how it is set up, and copy what makes sense of its script settings. Basically you’ll need to run multimarkdown first, to get a .tex file, and then use latexmk to handle the typesetting to PDF.