Ah yes, welcome to InDesign translation workflows! You will always need to relink all of your images.

(If you started with embedded images, you unfortunately just learned why you must un-embed images before starting a translation.)

Nice to see someone in the Adobe forums actually using TM tools.

I'll answer your questions, but not in the order you wanted. Sorry, don't know how much you know about ID, and I don't want to give you any misleading advice. There are two possibilities here:

1) This document was built the right way. The images are in separate linked files. InDesign will display a low-res preview that gets saved in the INDD. The high-res images (Photoshop files, TIFFs, whatever) are separate files.

2) This document was not built the right way. The high-res images were pasted directly into InDesign.

So, in order to give you the right suggestions, you need to visit your Links panel so we can figure out if your images are linked or embedded. Go to Window -> Links. What do you see? What version of ID are you using? Do you see the names of missing files in that panel? Do they have the red-circle-with-question-mark?

... Sorry, you're probably out of luck.  "Doing it right" would include either a) giving you all the tools necessary to complete your task, or b) giving you instructions "just send us back the IDMLs and we'll relink the images." Here's what is going on here: When the comics-layout guy placed the Page 6 illustration into InDesign, InDesign harvested a low-res 72ppi preview from the probably-300-ppi image file, and saved it in the INDD. That's what you saw when you opened the source file. When you exported IDML, you converted that INDD into a markup file - IDML - that has pretty much everything in it that the INDD did, barring the image previews. That's why, when you open up your IDML, you just get the grey boxes -the IDML knows where those images were, but it doesn't have the preview data anymore.

What I would guess that you want, here, is to see your translations in context with the images. Do you need to manipulate speech bubbles to account for text expansion? If so, make a PDF of the client's source doc and place on a new layer in ID, in front of the missing images but behind your translation. That way you can see what you're doing.  Are you trying to deliver something that looks exactly like what you received but for the translated content? I think that the only way to get that will be to copy/paste your translated content back into the source file, once you've manually stripped out all of the source language. That will take you rather more than one click. 

I don't know if your customer/supplier/whathaveyou is CAT-savvy enough to understand why the low-res previews are blown out of the translated doc, otherwise I'd advise you to just deliver the IDML. Chances are slim, right?

Sorry to be the bearer of bad news. If you can fill in any more details, I (or one of the other regulars) might be able to make a more useful suggestion. If this is going to happen to you on a regular basis, then the closest thing to your one-click solution would be to prepare your files carefully before starting translation by placing all to-be-translated content on a separate layer, and to have a Javascript set up to automatically take all of the content from one file (your post-TM IDML that you just opened in ID) and spit it into another file (your client's source file, stripped of all source language so it's just the 72ppi image previews). That's... uh... sorry, that's still lots of clicks. I've found over the years that translation workflows tend to be click-heavy unless everyone involved understands translation workflows. If your client is a savvy translation consumer, then you can email them with your concerns ("I can't give you final files that have your image previews still in 'em because I am a TOTAL PRO and am using computer-aided translation tools that only work on IDMLs which strip out image previews") and their response will be "Oh yes translation provider, let's adjust our workflow so that it is functional for all participants." Or "Heck, just send the IDMLs and we'll take it from here."