I’m pretty sure all the ones that take place in custom locations like Olivia’s hideout, Diana’s cabin, etc are pre-rendered particularly since that would be vastly more efficient than loading in all those assets to work only with fixed camera angles.
That whole scene for instance would be an utter waste given the player never goes near it and there’s a decent chance that, knowing they were going to use depth of field - they never properly detailed half the assets involved.
I suspect the issues are less about the compression and more about them using colour balancing adjustments to compensate issues with using the Glacier engine to do the rendering and with zooming in so close to the hero models.
For example, this one they’ve washed out the blacks a little and due to the colour pallette if you ask Photoshop to autofix it, the main response is to put a weird blue wash over it.
But if I use curves to sharpen it to cinematic blacks, 47 and Grey suddenly look too sharp - they look less like men in the dessert and more like a pair of mannequins riding in a car at night.
Lights and shadows are super hard to get right, arguably one of the hardest and most processor intensive parts (next to flexible polygons) and the models were generally designed for the kind of mid-range lighting that we have in a Hitman game in general.
This is part of why the in game cinematics tends to use every excuse to throw a big lighting effect and in and put heavy filters on “footage” to make it more of a suggestion of the movie than the actual movie per se.
Heavily stylized, but in a way you expect - is much easier to accomplish passably than trying to work out a way to do proper fidelity of a video in a video.
Similarly if we go to the most washed out scene:
Looks foggy etc, but if you tweak it up closer to how it renders naturally in game - Grey’s wardrobe makes him start to look like a floating head in the scenery and 47’s beautiful warm head becomes an eye magnet:
The original footage probably went through multiple drafts of colour correction and filters to get the best effect, but you can see that when you shift it you can change the whole feel of the shot and that the more you shift, the more resolution you lose and “compression” appears.
In a regular movie or with a big budget they’d have options to do things like have Grey change suits, put more sophisticated lighting setups to minimize the need for colour correction etc - but given that the cinematics are a tiny portion of the average player’s experience time, it’d be very hard to justify that.
There is also the matter that, for example, the model of Lucas Grey can be rendered to exceptionally high fidelity, and it looks fantastic as a still.
However if you have that character in animation and you can see him to that level of fidelity it starts to raise the level of animation fidelity required to avoid the uncanny valley effect. You see him rendered that well you expect him to breathe, you expect him to blink, you expect his hair to sway a little if he’s outdoors where there could be a breeze, you expect microexpressions, etc.
If you don’t get those then you start to feel like there’s something wrong, and that would contribute more to the feeling of Lucas as a dread antagonist rather than as 47’s long lost brother. There’s also the issue that frequently you have a limited range of motion etc that a character rendered in say, a coat, can do before it starts to become… well hilarious.
