As long as we're on the subject of Laserschwert, did you know that he recently gave his frame-ready poster download of the Monkey Island 2 cover art a material upgrade? You better give your eyeballs a head-up on this one.
You see, a 1991 issue of the UK-based magazine "The One" contained a large, fold-out poster of Purcell's artwork, making it a superior scanning source to the box, or for that matter anything else known to exist. A generous fan in possession of that particular issue heard the call of duty and sent the precious reference material to the pre-eminent authority in this field, who then worked his arts:
Read more and collect your upgrade in Laserschwert's stickied poster thread, where the latest version of his offerings can always be found.
Boy, it's almost like reading the forums is a good idea or something.
Source: The Forums
I had the new scan (A) and my old restoration (B). Using an image distortion tool I deformed A to be perfectly aligned on top of B. Then I picked out several parts of the artwork that could be found on both images (i.e. the parts that weren't covered by logos) to train my AI model on.
"Training" basically means that the AI tool tries to learn how it needs to change the pixels of image B to look like image A. This is an iterative process, so I usually let it run through each night (when I don't need my PC), improving the model for a few nights (I think I let this one run for a total of 20 or 30 hours). In theory this would at some point lead to a model that would make image B look exactly like image A (at least the parts I trained it on), but obviously you are only approaching that point and never fully reach it. But it doesn't really matter, because I don't need my whole image B to turn into image A (after all I have a huge portion of image A already intact - the new scan that I used for training). What I DO want is to have all the other areas of image B (the ones that are covered by logos on image A) turn into something that might look like it could be part of image A. I wouldn't use the entire artificially generated image, only the parts I need to remove logos and extend the edges.
Combining those two is not done automatically, but regular Photoshop retouch work.
It’s all wildly impressive stuff whatever the case, and the results speak for themselves. I already thought they looked great, but it’s like I’ve put some glasses on I never realised I needed.