Gamers react with overwhelming disgust to DLSS 5’s generative AI glow-ups

WereCatf

Ars Tribunus Militum
2,830
I wanna join in on the fun:
1773765174531.png
 
Upvote
620 (633 / -13)

Sarty

Ars Tribunus Angusticlavius
7,816
I'm actually pretty meh on the Resident Evil example that headlines the article. If you'd told me the real one was the DLSS-5 one, and the DLSS-5 one was the real one, I'd have believed you, and in a vacuum, I'm not really sure I have a strong preference between them.

But we're not in a vacuum. One of them is the finished product of a creative team's art direction, and the other is a slop machine's interpretation of how the creative team should have made it look. I don't care what the pile of soggy linear algebra "thinks" about art direction, at all.
 
Upvote
307 (340 / -33)

Varste

Ars Praetorian
534
Subscriptor
DLSS is one of the few recent bits of tech I was genuinely excited for. The things it could do even in the 2.0 iteration were impressive to me, and since I'm usually quite behind on PC hardware, felt like a real promising thing.
And now this. Nvidia is so far up their ass on AI slop that this was, I suppose, inevitable. Not just the characters looking like that braindead "hire fans lol" tweet about Aloy from Horizon not being "pretty" enough; as briefly mentioned in the article, shadows are gone. I think more stress should be put on just how bland and flat it makes environments, which is just as bad as the character side of things. While this is small potatoes in the current timeline of "AI ruining things", I cannot wait for the bubble to pop sooner than later. Preferrably before all these stupid data centers get built and then abandoned.
 
Upvote
245 (265 / -20)

wrylachlan

Ars Legatus Legionis
14,870
Subscriptor
I mean… maybe I’m just a basic dude but the DLSS5 versions look pretty great to me. I’m not getting uncanny valley so much as much higher detailed VFX. To the extent that the more detailed versions stray from the artists vision, then sure, that’s not good - the details should help tell the story. But if it’s just a computationally cheaper and easier to implement way of arriving at a level of detail that the artist wants then it’s fine. I predict this is going to be a tempest in a teacup that goes away when cards and games that actually implement this are released.
 
Upvote
123 (314 / -191)

ERIFNOMI

Ars Legatus Legionis
17,192
I'm actually pretty meh on the Resident Evil example that headlines the article. If you'd told me the real one was the DLSS-5 one, and the DLSS-5 one was the real one, I'd have believed you, and in a vacuum, I'm not really sure I have a strong preference between them.

But we're not in a vacuum. One of them is the finished product of a creative team's art direction, and the other is a slop machine's interpretation of how the creative team should have made it look. I don't care what the pile of soggy linear algebra "thinks" about art direction, at all.
Exactly. Some of the examples don't really look worse necessarily, at least in the very limited side-by-side comparisons we have, but they do look different. The main RE still that they keep showing off looks like a completely different mood. The original looks like an awful, shitty, rainy day and I bet that fits the tone of a RE game better than the afternoon summer shower look the DLSS5 version gives. You can't just put more random lights in a scene and make it "better."
 
Upvote
254 (268 / -14)

Frodo Douchebaggins

Ars Legatus Legionis
11,995
Subscriptor
while Digital Foundry described the “transformational lighting” effects as “astonishing” numerous times in its write-up, the reaction from the rest of the gaming world has been overwhelmingly negative so far.


"astonishing" isn't always positive. But yeah…
 
Upvote
103 (105 / -2)

ERIFNOMI

Ars Legatus Legionis
17,192
I mean… maybe I’m just a basic dude but the DLSS5 versions look pretty great to me. I’m not getting uncanny valley so much as much higher detailed VFX. To the extent that the more detailed versions stray from the artists vision, then sure, that’s not good - the details should help tell the story. But if it’s just a computationally cheaper and easier to implement way of arriving at a level of detail that the artist wants then it’s fine. I predict this is going to be a tempest in a teacup that goes away when cards and games that actually implement this are released.
There are some things that are objectively better. In the RE example, her hair self-occludes which looks pretty good, though maybe a bit overkill here because it's looking more like she missed her last appointment to touch up her roots. The skin on her fingers look like they have some subsurface scattering to make them look less like a flat skin texture.

But everything in the background has been given so much extra light and everything is shinier for some reason.

And if you watch the actual video demos that show movement, characters usually have a glowing aura around them. Like it takes multiple frames to get enough information on what was previously obscured by the character to nail down the colors and before that they overshoot to something too bright and lacking detail.
 
Upvote
212 (215 / -3)

Carewolf

Ars Legatus Legionis
10,364
I mean… maybe I’m just a basic dude but the DLSS5 versions look pretty great to me. I’m not getting uncanny valley so much as much higher detailed VFX. To the extent that the more detailed versions stray from the artists vision, then sure, that’s not good - the details should help tell the story. But if it’s just a computationally cheaper and easier to implement way of arriving at a level of detail that the artist wants then it’s fine. I predict this is going to be a tempest in a teacup that goes away when cards and games that actually implement this are released.
It is a slop filter, makes the game look like AI generated slop. Great if you like the slop look, but it is not exactly the zeitgeist at the moment.
 
Upvote
166 (232 / -66)
Post content hidden for low score. Show…
I mean… maybe I’m just a basic dude but the DLSS5 versions look pretty great to me.
Um wow, no accounting for taste I guess.
But if it’s just a computationally cheaper and easier to implement way of arriving at a level of detail that the artist wants then it’s fine.
This is very much not what it is.
 
Upvote
-49 (104 / -153)
Post content hidden for low score. Show…
Post content hidden for low score. Show…

Sajuuk

Ars Legatus Legionis
12,855
Subscriptor++
As always, multiple things can be true.
  1. It is technically impressive
  2. It does flatten any distinct art style into a generic template (and really fucks with lighting, based on breakdowns I’ve seen)
  3. It really does look like a literal “Gamer chud sexy sex filter.”
Seeing Digital Foundry glaze it until their eyes rolled into the back of their heads was deeply disappointing.

Edit: be, not bed.
 
Last edited:
Upvote
346 (363 / -17)

Mechjaz

Ars Praefectus
3,262
Subscriptor++
There are some things that are objectively better. In the RE example, her hair self-occludes which looks pretty good, though maybe a bit overkill here because it's looking more like she missed her last appointment to touch up her roots. The skin on her fingers look like they have some subsurface scattering to make them look less like a flat skin texture.

But everything in the background has been given so much extra light and everything is shinier for some reason.

And if you watch the actual video demos that show movement, characters usually have a glowing aura around them. Like it takes multiple frames to get enough information on what was previously obscured by the character to nail down the colors and before that they overshoot to something too bright and lacking detail.
Ambient occlusion already has this problem in the negative: there's an aura of shadows around things, even when the thing isn't casting a shadow in itself. An easy example is the protagonist in GTA V against a bright background, where there's always a subtle, dark haze around the character. Of course it manifests in other ways, but that's a pretty clear one.

It's a bummer that instead of working on this and other actual rendering issues, Nvidia is working on making sure that everyone looks generically fuckable.
 
Upvote
144 (148 / -4)

Wilkey

Smack-Fu Master, in training
77
I mean… maybe I’m just a basic dude but the DLSS5 versions look pretty great to me. I’m not getting uncanny valley so much as much higher detailed VFX. To the extent that the more detailed versions stray from the artists vision, then sure, that’s not good - the details should help tell the story. But if it’s just a computationally cheaper and easier to implement way of arriving at a level of detail that the artist wants then it’s fine. I predict this is going to be a tempest in a teacup that goes away when cards and games that actually implement this are released.
That's fair and I agree on all points.

I have far less of an issue with the appearance of the foreground character in an absolute sense than I do with the loss of atmospheric haze at the background umbrella-carrying characters. That, for me, is the "improvement" that significantly alters the feel of the original in a harsh, detracting manner.

Nvidia's first mistake was in re-rendering an already existing property that people know and have opinions about. Their second mistake was comparing DLSS-5 against "off" instead of DLSS 4 or 4.5.
 
Upvote
88 (96 / -8)

jm_leviathan

Ars Scholae Palatinae
940
I'll leave it to more knowledgeable and passionate folk to address the practical and philosophical issues here. But even just as a branding exercise, this is clearly not great. From the beginning, DLSS has fought an uphill battle to gain acceptance from gamers and seemed to have finally achieved that (albeit still with some resistance on the frame generation side of things). By calling this DLSS 5, it risks tainting the broader enterprise: "Oh yeah DLSS is pretty great... no, not that DLSS."
 
Last edited:
Upvote
102 (106 / -4)
Post content hidden for low score. Show…

Missing_Linc

Smack-Fu Master, in training
23
I find it pretty telling that Todd Howard was really excited by it. Bethesda now don't need to do anything to improve their graphics, they can just let Nvidia handle it all for ES6.

I did find it funny that the demo was rendered on 2 5090s considering the recent pricing hikes and availability issues.
 
Upvote
128 (131 / -3)
Post content hidden for low score. Show…

WereCatf

Ars Tribunus Militum
2,830
All On examples are objectively better looking than the Off examples. I get that lots of people hate AI for understandably good reasons, but this isn't one. It certainly isn't "AI slop"
I guess it depends on one's taste and how much one's pants snake rears its head when looking at AI-generated female faces. All the examples I've seen drastically alter the games' aesthetics and make everything look horribly samey and in many cases absolutely demolishes the characters, like e.g. in the Hogwarts-game, one of the boys who is supposed to be a teen ends up looking like a 40-year old man and an elderly lady looks like a fricking mummy -- it doesn't look anything even remotely close to how the game is supposed to look.

You do you, but I, for one, am not excited about every game having a generic AI-look to them.
 
Upvote
114 (135 / -21)
I find the most irredeemable part (and likely impossible to be solved by this approach) is just how terribly it messes up the lighting. Remember, the stills and clips released for this are the absolute cream of the crop, best looking examples they could come up with after assigning a full marketing department to work on it for months. Despite that, these are the best they could find.

Lol at the RE:R still and tell me where the light sources are in the original versus the DLSS. Lighting in the original is far from perfect, but it at least makes some sense based on the environmental cues like the illuminated sky. Items in the background cast shadows and have depth. The DLSS one appears like the character model is placed in front of a far too sharp 2d backdrop then photographed by a camera with the flash on.

This is the best it can get and if both images were submitted in a middle school art class, the original might get an A- while the "improved" version would be lucky to get a C.
 
Upvote
131 (143 / -12)
Post content hidden for low score. Show…

PYR0DR490N

Smack-Fu Master, in training
74
I mean… maybe I’m just a basic dude but the DLSS5 versions look pretty great to me. I’m not getting uncanny valley so much as much higher detailed VFX. To the extent that the more detailed versions stray from the artists vision, then sure, that’s not good - the details should help tell the story. But if it’s just a computationally cheaper and easier to implement way of arriving at a level of detail that the artist wants then it’s fine. I predict this is going to be a tempest in a teacup that goes away when cards and games that actually implement this are released.
Computationally easier?? Really?
In these demos, DLSS5 uses an ENTIRE_RTX5090 just to itself! How is that possibly "computationally easier"?
 
Upvote
255 (265 / -10)

jhodge

Ars Tribunus Angusticlavius
8,661
Subscriptor++
If NV is being truthful about DLSS5 being another tool for devs to use as they see fit, I don't see any problem with it. OTOH, altering the look of the game after the fact should be out of bounds. Then again, on the third hand, gamers have been modding games with "upgraded" textures and ReShade forever, so it might not be unreasonable to allow for user-directed toggling of the effects.

IMO, it's all about ensuring that the default presentation is the one the game developers intended.
 
Upvote
25 (45 / -20)