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:
However, they appear to only ever hint at the fact that character faces are being "enhanced" in extremely strange and biased ways, especially female characters. The results appear similar to GenAI pictures from prompts like "make X look pretty/handsome", adding what appears to be make-up effects rather than actually simulating a more realistic lighting behavior.
What are everyone's thoughts? And how do you think the rest of the DF crew will respond to this?
Maybe this is the tech that finally fixes Aloy's face? Jokes aside, I think it's hit and miss, I actually like the Grace example but found some of the environmental changes in other games less convincing since they brighten up too much for my liking.
I could see it being useful for certain things, like landscapes or photorealistic player likenesses in sports games like NBA 2K and FIFA, but I don't like how it changes the faces of characters in ways that the artists and developers surely never intended.
DLSS has always been about neural rendering. Render less and get a compute-based uplift. Sudden change like the facial rendering is certainly shocking but it's all about lowering the total compute cost for making images. In there is where some great benefits live like we've seen from DLSS in the past wrt upscaling, anti aliasing and inter-frame generation. "Niceifying" an image like the stones and landscapes as well as tonally balancing the image is powerful and I'm assuming a bunch of control surfaces will help creators tune that. The facial work is dramatic and probably not many of us will read the documentation on what controls are there to tune - the rest will probably second-guess the creator's intent but that's how it goes.
I've been following DLSS since the beginning and love that there's neural rendering techniques to improve clarity, detail, surface and material stability, frame rates and now more holistic uplifts.
@Knightfall I doubt NVIDIA would use marketing material without Capcom's or Bethesda's consent.
Maybe the original game's artists even worked on it, I cannot say in this case, but in the future they will, so it will be artistic intent then.
Just consider it an additional graphics setting, and if you don't like it go back to the default.
They won't be able to force it upon you, since it seems to be rather power hungry.
I definitely think whatever material its trained on to represent human faces is absolutely not representative of actual people's faces, and more like some kind of weird amalgamation of TV and film appearances.
The reaction to this appears to have been pretty toxic and it does not really seem to relate to the tech or output itself. It seems to be part of the overall sentiment amongst a small but very vocal minority that anything "AI" is bad and they will rant about it aggressively if anyone says anything positive. Almost everyone against seems to be using the same buzzwords, the same anti-AI cliches. I am not for or against AI, it is neither good nor bad, it is a tool, nothing more, but too many people seem to have lost their objectivity and attack it's use without thinking.
@Ukigumo Yeah, that's a fair point. If it's being shown in this marketing video, it makes sense to assume/hope they had the go-ahead from the developers.
It's interesting tech. I was an art director, so seeing this in action at first rubbed me the wrong way, but the more I think about it, the more I can see its paradigm shifting potential.
it seems like the main 2 photos are extra touched somehow. The fact is she looks like Scarlett Johansson with DLSS5. When I use the slider it seems to also be in a different frame in time. The woman's eyebrow is lifted and the lips are different. In many of the other examples it seems to be just lighting.
I think NVIDIA messed up with those 2 pictures.
As for DF it just looks bad 2 people with pixel trained eyesight didn't see Scarlett in there.
As for the backlash the internet does what the internet does.
If I was an Art Director working in games I'd feel pretty depressed today. In fact, looking at the imagery DF has posted, I'm still pretty depressed today!
My personal feeling is that DLSS has gone from "getting rid of the jaggies", which I do want, to something that I don't ever want under any circumstances. Just because you can do something, doesn't mean you should. It's going to be interesting to see how Nvidia deals with the backlash.
We've seen recently the transformative effect that things like path tracing and denoising can have on how a game appears. I was as sceptical as the next person when I saw the RE: Requiem comparisons, but only because it's such a massive leap. If the base geometry and textures aren't altered, and it is just enhanced lighting, then Nvidia would do well to show that to people as soon as they can to head off this "AI slop" perception.
Also find it funny that they use basically the worst Grace image to compare against the DLSS 5 one: the path traced one is slightly less jarring.
@Claw Nvidia has gone on record to say that developers are able to "mask" objects/models to which the enhancements should not apply, so it's likely that actual application of DLSS 5 would be done in a way that doesn't produce results that radically differ from overall art direction. I do think the demo had other issues as well, though, the AC Shadows landscapes were consistently looking like a downgrade when DLSS 5 was enabled.
@Hustler_One I've also spotted something interesting on Nvidia's own page promoting DLSS 5 - it has a visible effect on environments too. Up until now I think I've been so distracted by the alarming changes to character faces, that I completely missed the other "enhancements". Take a look at this, and examine the background: https://www.nvidia.com/en-us/geforce/news/dlss5-breakthrough-...
If I was the environment or lighting artist that worked on this scene, I'd probably be a bit hacked off.
I definitely think whatever material its trained on to represent human faces is absolutely not representative of actual people's faces, and more like some kind of weird amalgamation of TV and film appearances.
I think you're right — besides the glaring example with Grace (where DLSS5 appeared to extrapolate cosmetics like lip color & facial contouring) character faces in other examples were lit in ways that didn't fit the scene and suggested that the model was targeting a "ring-lit" effect.
The thing with video-games is you're constrained by the technology. Between what is possible and what is reasonable on prevailing hardware - chances are compromises will be made regarding one's vision for the look of a given title.
I didn't work in 3D, most of what I did was 2D poster and print ad work. In the creation of such, to achieve my vision, I would often lean on filters, stock imagery and other such to craft images, shots and effects I wouldn't have the time or ability to execute otherwise. It might not have fit my initial vision 100%, but perfect is the enemy of good.
Another of the tricks I often employed to mask the fact that an image was heavily composited was to overlay a film grain texture. It would modify the colour of the image, sometimes requiring tweaking, sometimes I'd prefer the modification, but regardless, the grain sold the image as a cohesive unit.
I don't think this tech is all that far away from that idea. And if you don't like it, switch it off.
Forums
Topic: DF's coverage of DLSS 5
Posts 1 to 19 of 19
Please login or sign up to reply to this topic