Why Red Alert's cutscenes are so bad to compare with Tiberium ones?

[Niragiです]

[Nyerguds]

The cutscenes were enhanced using AI upscaling from an original resolution of a mere 320x176. For more information on that, see Jim Vessella's video on the subject:

https://www.youtube.com/watch?v=ikJLYYTrIxs

As for the difference, well, the Tiberian Dawn briefing cutscenes are always one-on-one, while in Red Alert, they are generally a briefing room with about three people in there. This means the faces of the people in the scene are much smaller in the overall image.

AI upscaling can do amazing things, but of course, the process works a lot better if you have 130 pixels of face height than if you got just 40.

[Niragiです]

Originally posted by Nyerguds:

The cutscenes were enhanced using AI upscaling from an original resolution of a mere 320x176. For more information on that, see Jim Vessella's video on the subject:

https://www.youtube.com/watch?v=ikJLYYTrIxs

As for the difference, well, the Tiberian Dawn briefing cutscenes are always one-on-one, while in Red Alert, they are generally a briefing room with about three people in there. This means the faces of the people in the scene are much smaller in the overall image.

AI upscaling can do amazing things, but of course, the process works a lot better if you have 130 pixels of face height than if you got just 40.

It's pity they didn't save original high quality source videos from cameras

[Nyerguds]

Note that the expansion videos of Red Alert are better quality; the briefings there are once again one-on-one, and the full video frame in the original video material is 4:3 full screen 320x240 resolution.

Originally posted by Niragi:

It's pity they didn't save original high quality source videos from cameras

They do have all of that material of C&C1, but those are the originally recorded VHS tapes, not digitised versions. And those tapes seriously degraded over the last 25 years, to the point that the AI upscaling on the half-sized digital cutscenes generally looks better.

Not to mention, these are greenscreen videos so they'd still need all the post-processing redone. And they didn't recover any of the digital assets used to make the environments that filled in the greenscreen.

[Niragiです]

Originally posted by Nyerguds:

Note that the expansion videos of Red Alert are better quality; the briefings there are once again one-on-one, and the full video frame in the original video material is 4:3 full screen 320x240 resolution.

Originally posted by Niragi:

It's pity they didn't save original high quality source videos from cameras

They do have all of that material of C&C1, but those are the originally recorded VHS tapes, not digitised versions. And those tapes seriously degraded over the last 25 years, to the point that the AI upscaling on the half-sized digital cutscenes generally looks better.

Not to mention, these are greenscreen videos so they'd still need all the post-processing redone. And they didn't recover any of the digital assets used to make the environments that filled in the greenscreen.

considering huge neural networks have developed this shouldn't be big problem to remaster those cutscenes again, those legendary Stalin, Nadia, Gradenko, Kukov cutscenes still look awful

[Nyerguds]

AI has made huge leaps in the past years... but this game was released 3 years ago, you know. A lot of that development was after it was released.

And 40 pixels is still just 40 pixels. Unless you go and train an AI specifically to reconstruct these specific faces, it's never going to be much better than that. There's just not enough data on the screen.

[Niragiです]

Originally posted by Nyerguds:

AI has made huge leaps in the past years... but this game was released 3 years ago, you know. A lot of that development was after it was released.

And 40 pixels is still just 40 pixels. Unless you go and train an AI specifically to reconstruct these specific faces, it's never going to be much better than that. There's just not enough data on the screen.

just picking actors' photos would be quite enough data for AI
R.I.P. Eugene Dynarski

[Nyerguds]

Originally posted by Niragi:

just picking actors' photos would be quite enough data for AI

I don't think there is technology yet to specifically tell an AI, in every scene, which blob is supposed to be which actor, and give them higher res image material to help reconstruct it. And even if that tech exists, it definitely didn't three years ago. And they're not going to re-do it all now.

I mean, feel free to experiment with AI yourself to see if you can get better results... but I doubt you're actually going to publish a 20gb mod to enhance it all.

[edit]

For the record, they sourced the video material for the Remaster from the Playstation versions where available, both because it's in higher colour, and because, unlike the PC version, it's in correct 4:3 aspect ratio there, which also makes the videos slightly larger (320x176 rather than 320x156)

Here's an example showing the smoother colour fades on the PSX video material:
https://i.imgur.com/FeGX6NF.png

[tacticalnuke]

Originally posted by Nyerguds:

Originally posted by Niragi:

just picking actors' photos would be quite enough data for AI

I don't think there is technology yet to specifically tell an AI, in every scene, which blob is supposed to be which actor, and give them higher res image material to help reconstruct it. And even if that tech exists, it definitely didn't three years ago. And they're not going to re-do it all now.

I mean, feel free to experiment with AI yourself to see if you can get better results... but I doubt you're actually going to publish a 20gb mod to enhance it all.

[edit]

For the record, they sourced the video material for the Remaster from the Playstation versions where available, both because it's in higher colour, and because, unlike the PC version, it's in correct 4:3 aspect ratio there, which also makes the videos slightly larger (320x176 rather than 320x156)

Here's an example showing the smoother colour fades on the PSX video material:
https://i.imgur.com/FeGX6NF.png

Surprised they didn't re-shoot the fmv's for the remaster.

[Cat]

Originally posted by tacticalnuke:

Originally posted by Nyerguds:

I don't think there is technology yet to specifically tell an AI, in every scene, which blob is supposed to be which actor, and give them higher res image material to help reconstruct it. And even if that tech exists, it definitely didn't three years ago. And they're not going to re-do it all now.

I mean, feel free to experiment with AI yourself to see if you can get better results... but I doubt you're actually going to publish a 20gb mod to enhance it all.

[edit]

For the record, they sourced the video material for the Remaster from the Playstation versions where available, both because it's in higher colour, and because, unlike the PC version, it's in correct 4:3 aspect ratio there, which also makes the videos slightly larger (320x176 rather than 320x156)

Here's an example showing the smoother colour fades on the PSX video material:
https://i.imgur.com/FeGX6NF.png

Surprised they didn't re-shoot the fmv's for the remaster.

1. There would be no point to reshoot.
2. Some of the actors passed away or have aged.
3. That's throwing money away for no reason.

[tacticalnuke]

Originally posted by Cat:

Originally posted by tacticalnuke:

Surprised they didn't re-shoot the fmv's for the remaster.

1. There would be no point to reshoot.
2. Some of the actors passed away or have aged.
3. That's throwing money away for no reason.

ᚨ. Im talking about the CGI stuff that's aged incredibly badly. The videos with the actors seem to have worked well though. Just to clear that up ;)

ᛒ. ᚷᛁᚢᛖ ᚺᛁᛗ ᛏᛁᛗᛖ ᛏᛟ ᚠᛁᚷᚢᚱᛖ ᚦᛁᛊ ᛟᚢᛏ ;)

[Nyerguds]

The CGI stuff actually upscales pretty well with AI. I had some really good results with that, using the ESRGAN AI.
https://www.reddit.com/r/GameUpscale/comments/aiatem/command_conquer_1_red_alert_1_renders_with_esrgan/