@AndyGilleand In some examples it is what you describe and it looks better but also worse (ring-light effect on faces).
But in the Grace image.
Left image is Grace, right is Scarlett Johansson. The entire internet can see, it is not misinterpretation.
They're both Grace. One is Grace with flatter shading, without self shadowing on the micro detail, without improved subsurface scattering, with less realistic specularity, etc. It's exactly the same model, exactly the same textures, just rendered in a more natural way. The weaker rendering made her not look quite right in the original game. The game engine wasn't living up to the original artwork.
Strangely, Alex himself references how DLSS 5 is affecting details which are not from the models/textures but most likely from some external dataset (and he directly references the make-up issue). But some posters here would have him check his own eyes first 🤡
Alex is wrong. There's no added makeup, no added details. Again, the previously referenced video shows that very clearly that it's all the same existing detail. The main difference is that normal maps only create slight variations in shading, whereas the AI can interpret that as if it were very high polygons and create tiny shadows / AO from that micro detail on the surface, which makes those details stand out much more than normal maps do. Because they stand out much more, they can seem like they're something that wasn't in the original image, but when you actually compare it closely side by side, you can see those details were all there in the original render.
Again, I can completely understand why people might THINK those things are true, but those initial assumptions don't hold up when you look closely at the original details. There's no added makeup etc. People may think it's silly, but yes, your eyes can in fact deceive you when you don't analyze things more closely.
I think I trust my eyes (and the mountains of evidence, and DF's own written piece on the QA which mentions details being generated, and the opinions of several game developers, and Nvidia's own words about DLSS 5) over some nobody YT channel (who even is whizz dumb lol) and all the doubling down on easily disproven arguments.
I have seen nobody give any actual evidence for claims of generated details (which Nvidia did NOT say it does) and regardless of a youtuber being a nobody, they give actual objective evidence that is far more reliable than "trusting my own eyes". Our eyes are notoriously untrustworthy when it comes to technical analysis.
Considering we're on DF, surely actual objective technical analysis would be more valuable than subjective gut impressions, no?
Considering we're on DF, surely actual objective technical analysis would be more valuable than subjective gut impressions, no?
TBH, I'm disappointed with DF and this video. They bowed down to the mob and the power/fear of the YT algorithm.
..and I don't like JL's subtle digs at the team, on Bluesky, over the original video and the way he's positioning himself as the 'Hero of the People' over the whole issue and the fact he wouldn't appear in this video seems like it was his way of washing his hands of the whole affair.
If he had such a problem with the original video, he should have kept that between the team, in private.
Airing ones dirty washing in public is not a good look for DF and those who do it.
So Digital Foundry's coverage of DLSS 5 is being questioned around the web, or more precisely, people are having a much more negative reaction to the visuals produced by DLSS 5 than Rich and Oliver had in the video piece. They had a very positive tone on their video piece as well as their written article:
It's been a few days now. We've all had some time to digest and the thing that's leaving a bad taste in my mouth is mostly what you illustrated in your original post. Yesterday I watched the 2 videos DF put out back to back and the contrast is jarring. In the original piece both Richard and Oliver were genuinely excited and looked almost childlike in their enthusiasm. Which I liked a lot. It rubbed off on me as the viewer. I really got the sense I was looking at something spectacular.
Then the backlash happened and that second video was put out and the tone was much different. That even has an apology in the title itself "We should have waited". Ok....wait for what exactly? For the pitchfork hordes to tell you what to feel and think?
That kind of conforming to the masses annoys the ever living hell out of me. They weren't the only ones by the way. Daniel Owen did the exact same thing. Even going as far as posting a follow up vid with a thumbnail mocking Jensen Huang. In the case of DF I can't shake the feeling Alex was flown in for damage control.
And it's not a first either. This is the second time that I can recall seeing DF fold under pressure. The first time being back in the day during the JK Rowling controversy where they initially completely ignored the existence of Hogwarts Legacy. That wasn't a good look and neither is this.
Even though most people will never admit it they respect people who stand their ground, even if they don't agree with them. Never let fear of the masses be your guide because it will absolutely cost you your integrity. Because let's be real. Going forward with DLSS5 coverage, what will I be watching? Their honest opinions, or what they think the pitchfork horde wants to hear?
History doesn't remember the people pleasers, it remembers those who voiced their opinion regardless of consequence.
I agree with the last two posts. It feels like they're letting social media influence their perspective, which is never a good idea. It's impossible to gauge actual consensus by looking at social media. Algorithms promote comments that drive engagement. A large amount of the time, these are comments that focus on controversy. Because that stuff is controversial and generates a lot of discussion, that inherently tends to mean that it's not the majority opinion. Honestly, the more you see of a loud opinion going viral on social media, the less likely it is going to be the most popular opinion out there.
However, without taking proper statistics that are accurately representative of the diversity of the population, you're never going to be able to make accurate assumptions about what most people are thinking. You will only be able to see what the loudest people are saying.
So it's best not to let those social media opinions influence your take on things. It can certainly be difficult to avoid that influence, but I'd hold DF to a higher standard than most people when it comes to that.
Every member of the DF team has its own opinion on the subject, obviously it is a very divisive subject. But I don't get the impression that Rich and Oliver turned coat and bowed down to the masses at all in the latest video. They bring more nuance to their views in a longer format but all of this was already present in their original video. The mood certainly changed (being shown new tech vs being at the center of a controversy on the internet will do that to you) but what I get is that they still think it is exciting new tech, it has potential and they're ready to cover it. And I respect that, I lean closer to Alex's view on this (both on an ethical and technical perspective) but I'll be watching how this evolves and how Nvidia itself reacts to the huge feedback it received.
Because let's be real, Nvidia has to find a way to make games look better and they won't achieve it with more raster. I was expecting neural rendering to be more about neural materials and whatnot, to have a progressive evolution along this line. And I was not ready for the video-to-video post processing that is going on in DLSS 5 with the model as it currently stands. Now if they are able to significantly improve it like DLSS 2 resolved the issues of DLSS 1, and we get an answer to all of our concerns on this tech, I can change my mind.
And I was not ready for the video-to-video post processing that is going on in DLSS 5 with the model as it currently stands.
To be clear, it's not a video-to-video model. It's a model that operates on several render layers, and is temporally consistent, but it's not converting video. It's applying a layer of light interactions on top of existing geometry and texture. It's effectively the same concept as ray reconstruction, except instead of running on an input that requires noisy raytrace samples, it can run without raytracing at all. Where it's a reconstruction method that "fills in the gaps" of what's missing in the lighting.
Video-to-video would take massively more hardware with massively more VRAM to run in real time, and honestly, it may not even be theoretically possible given physical limitations.
Because let's be real, Nvidia has to find a way to make games look better[.]
Do they? Take a look at their earnings, look at what percentage of their revenue comes from retail gaming customers versus every other aspect of their business.
@Frosh it's pretty clear that the goal of DLSS 5 for now is selling the idea for publishers that they can cut a lot of jobs related to art in their business. What's the point of a lighting or character artist when you can have "photorealism" as interpreted by some algorithm/dataset?
@Hustler_One Do some more research before you post such bold claims, as i have a feeling you wouldnt be able to back it up with facts.
If you want to follow your line of reasoning, then why do artists even let a game engine shade their model? Shouldnt they paint every single pixel? I thought we were past the days of sprites and C64.
Upon a closer look it is clear Grace's face is massively different, her lips are extruded, her eyebrows are higher, her eyes are looking in a different direction, her hand is transformed from plastic doll to human with extra detail and more. I'm not saying I am pro or against this, it's just new. Just the fact that in the story she's anxious to see if her father died is good enough to convince me it has messed up with the artist's intent.
About DF now and to the point, I think Oliver and Rich were indeed overexcited in the first video but I don't blame them. It's an impressive new technology they were honored to be the first to see, try and present to us. I would be excited too, I wouldn't be a bloody executioner to the people who invited me based on a hunch of a backlash, nobody would. Most of the other YouTubers rose their banners after the fact and from the comfort of their homes, that's not professional nor honorable, they just rode the backlash as wannabe AI slop crusaders. In a way they remind me of AI slop, fancy but cheap.
On the second DF video there are traces of Oliver and Rich trying to mend things according to the backlash but it's not a complete turnaround. What they are presenting is reasonable focus on the details, what still looks better and what should be toned down. It is as professional and honest as always. As it is the first time we hear Alex's opinion it is not fair to assume he "flew in for the rescue", he is just himself. Either way, DF didn't lose my faith overnight, I'm still subscribed and waiting for their next content as usual.
Upon a closer look it is clear Grace's face is massively different, her lips are extruded, her eyebrows are higher, her eyes are looking in a different direction, her hand is transformed from plastic doll to human with extra detail and more. I'm not saying I am pro or against this, it's just new.
From your post you strike me as a reasonable person, which is refreshing since having this debate on other platforms like the steam forums things descend into name calling and ad hominem arguments rather quickly.
Can I please invite you to watch just a few seconds of the video I posted earlier. The Grace comparison starts at exactly 0:37 seconds into the video and lasts just 10 or 20 seconds. If you would do me a favor and check that out and tell me what you think. You know what, here it is so you don't even have to look for it :
@Frosh it's pretty clear that the goal of DLSS 5 for now is selling the idea for publishers that they can cut a lot of jobs related to art in their business. What's the point of a lighting or character artist when you can have "photorealism" as interpreted by some algorithm/dataset?
The "algorithm" fully relies on that art. It doesn't create anything the artists haven't already set up. It also doesn't change the primary lighting. It only affects secondary lighting and the kind of light interactions with materials that modern game engines just can't simulate well. The underlying art is unchanged, which means you can't get rid of artists or the DLSS5 results will suffer as well.
On the case of Grace's face which is the most interesting (https://www.nvidia.com/en-us/geforce/news/dlss5-breakthrough-in-visual-fidelity-for-games/nvidia-dlss-5-resident-evil-requiem-geforce-rtx-comparison-screenshot-001/), it's unfortunate because the shots don't align perfectly (other shots do a better job in this regard). So that probably explains why her eyes are looking in a different direction.
Her shoulder on the right of the image (her left shoulder) is slightly higher, the opposite shoulder still higher but a bit less, but her face, if you look at the bottom of her chin to her hairline, her face appears to be the same size. That is very troubling to me because her lips are not. She appears to have bigger lips with DLSS 5 On in terms of geometry and this is the only shot that gives me this effect. I think what is called the cupid's bow appears more pronounced as well. So I don't know what's going on there. At first I thought it might be because the white roll (the white line on top of the upper lip) is lighted differently but there's more to it. It's really difficult to say with certainty that the geometry has not been altered there. Is this an artifact? Something the model is not supposed to do but did anyway? It's really hard to understand this technology, it's a bit of a black box to me.
On the case of Grace's face which is the most interesting (https://www.nvidia.com/en-us/geforce/news/dlss5-breakthrough-in-visual-fidelity-for-games/nvidia-dlss-5-resident-evil-requiem-geforce-rtx-comparison-screenshot-001/), it's unfortunate because the shots don't align perfectly (other shots do a better job in this regard). So that probably explains why her eyes are looking in a different direction.
Her shoulder on the right of the image (her left shoulder) is slightly higher, the opposite shoulder still higher but a bit less, but her face, if you look at the bottom of her chin to her hairline, her face appears to be the same size. That is very troubling to me because her lips are not. She appears to have bigger lips with DLSS 5 On in terms of geometry and this is the only shot that gives me this effect. I think what is called the cupid's bow appears more pronounced as well. So I don't know what's going on there. At first I thought it might be because the white roll (the white line on top of the upper lip) is lighted differently but there's more to it. It's really difficult to say with certainty that the geometry has not been altered there. Is this an artifact? Something the model is not supposed to do but did anyway? It's really hard to understand this technology, it's a bit of a black box to me.
No geometry has changed. The video above gives a good analysis of that. It's not a black box in terms of what the model is actually doing. It's a black box in terms of how the model makes its decisions. What the model is actually allowed to do is tightly controlled, because the base colors (including primary lighting), depth map, and motion vectors remain unchanged. What it's actually modifying is similar to the way ray reconstruction works. The same geometry, the same detail, just different shading of that detail. The main difference between DLSS5 and ray reconstruction is that it can take an input that doesn't even have any raytrace samples at all, making predictions about how pathtraced lighting would work in a scene with the existing objects and primary raster lighting.
There are times when dramatic changes in shading can make it appear like geometry has changed, but when you actually examine closely, it has not.
It's just unfortunate that the most controversial shot does not align because on this one (https://www.nvidia.com/en-us/geforce/news/dlss5-breakthrough-in-visual-fidelity-for-games/nvidia-dlss-5-resident-evil-requiem-geforce-rtx-comparison-screenshot-002/) we can see that the geometry is the same.
Let me just correct you though, the Grace I was talking about in your video is not at 0:37 but at 5:37. Your YouTuber agrees that this is the most controversial shot and indeed points out the extruded lips, eyes etc and it looks AI generated. So yeah nice catch.
Here's the correct link that will take you straight to the example instead of a Starfield NPC. Made it easy for you
So it seems DLSS 5 really is just some post-processing slop filter. It's all based on the actual image, so while it doesn't alter the geometry (there's no data for this) it will still make models and textures change radically once applied, which is why the changes can look not only awful, but also entirely inconsistent from one scene to the next. Hoping DF picks up on this and covers it accordingly.
Yeah, so as the model stands right now on this demo, it can definitely make stuff up (your typical generative AI hallucinations). Even though it is very constrained not to alter geometry, sometimes it will make mistakes. Since it is entirely generating the frame, it can add hair because on its training data the face had hair there and so it doesn't really make the difference between changing a texture and lighting. On the 'problematic' Grace, it is almost hallucinating teeth which were not there (it is very subtle I don't know if everyone will agree with me). And as Alex mentioned it hallucinated make-up like mascara.
Forums
Topic: DF's coverage of DLSS 5
Posts 41 to 60 of 94
Please login or sign up to reply to this topic