[Bandit LOAF]

ODVS has written a pretty fascinating article titled "Room with a View" all about the technologies he uses to help increase the resolution the gameflow assets from the original game. That's the part 'inside' the carriers: the backgrounds, the animated character 'gumps' and so on. They're all intended for a 640x480 screen even in the DVD version, so it's a real challenge to modernize them! Long story, short: ODVS isn't just pressing a 'make better' button on the original assets! He does a great job of walking us through everything... and sharing some pretty cool examples of his work! You can read the entire article here.




--
Original update published on August 23, 2026

 

[Wedge009]

I maintain that AI is a misleading term and that the models don’t think – but now that they can infer, they can recognise the nature of items and treat them accordingly.

Agreed. Unfortunately, the term AI seems to have gone the same way of 4K and such and we're stuck with it. Personally, in the context here, I prefer to call it machine-generated, but most people will still say 'AI' even though it really is just inference.

The Dekker picture is a remarkable recreation from the source material but even better is the Hawk and Panther comparison from the article. The generated example looks like it could plausibly scale down to the blurry source material, but the generator has no context of how the characters look in other scenes. The 'faked' example looks a lot closer to the characters we know, presumably because the tools are being given that context.

I cannot get over the sheer insanity of having to Deepfake someone’s own face back onto them to make them look like themselves, but you can’t argue with the results!

It is interesting that such tools are often used to produce something contrary to reality, but here is one case where we're actually trying to preserve reality. That does sound 'insane' indeed.

 

[ODVS]

Agreed. Unfortunately, the term AI seems to have gone the same way of 4K and such and we're stuck with it. Personally, in the context here, I prefer to call it machine-generated, but most people will still say 'AI' even though it really is just inference.

The Dekker picture is a remarkable recreation from the source material but even better is the Hawk and Panther comparison from the article. The generated example looks like it could plausibly scale down to the blurry source material, but the generator has no context of how the characters look in other scenes. The 'faked' example looks a lot closer to the characters we know, presumably because the tools are being given that context.


It is interesting that such tools are often used to produce something contrary to reality, but here is one case where we're actually trying to preserve reality. That does sound 'insane' indeed.

Yep, strong agree. We had so many other, more accurate terms six years ago (machine learning and deep learning, for example) that I really wish had stuck with the general population. But you just can't fight the cultural context of Hollywood making everone want their own Commander Data or HAL 9000. As soon as some prat used the term "AI", it was all downhill from there.

 

[cd1188]

@ODVS I'm using the same stuff like you 😄 but I give you a tip .. Starlight is bad for small faces.. try SeedVR2.5 ... but every software has pros and cons and you have to mix them together.

 

[kingmakeress]

Yep, strong agree. We had so many other, more accurate terms six years ago (machine learning and deep learning, for example) that I really wish had stuck with the general population.

True, AI is too generic a term and in many cases technically wrong.