DLSS 5 has huge problems, but it’s not the catastrophe it’s made out to be!

Furthermore, they claim that with this technology, it will be possible to achieve a level of graphics and visual effects that until now were only possible within Hollywood productions.

Everyone knows what kind of reactions this announcement caused. The most positive part of the whole story is that it has once again shown that gamers certainly have a good sense of humor. However, the point of this text is not to catch the "hate-train," but an attempt at a more objective look at the entire situation and an attempt to point out the good things that this technology will bring with it.

What exactly is DLSS 5?

To understand what exactly DLSS 5 is now, we must understand what it was before that. In its early stages (DLSS 1 & 2), DLSS presented itself as a technology aimed at improving game performance. This was done in a way that the game renders at a lower resolution – 1080p, and then the AI "stretches" it to a higher one – 4k. With DLSS 3 & 4, a step forward was made, and so we came to frame generation. In short, the AI takes two successive frames and analyzes the direction of movement of every object (not just the camera, but also particles, shadows, UI); in this way, a completely new frame is generated between the two frames.

DLSS 5 introduces a completely new concept called neural shading and moves away from a deterministic approach in favor of probability. Specifically, the old approach implied that if we want to see a reflection on something, the GPU must trace a light ray, calculate the angle of incidence, the surface texture, and where that ray will end up. It sounds like a lot of math, which it essentially is, and thus becomes quite expensive for the hardware.

Neural shading, on the other hand, involves introducing probability, so the GPU renders a "sketch" of the scene (low-res, basic shadows). That sketch is sent to a neural shader that doesn't calculate light, but predicts it. Additionally, the AI actively performs semantic segmentation, which involves identifying what is in the image, resulting in masking the scene in a way that it marks something as steel, something as leaves, something as skin, etc.

Nvidia actually wanted you to hate it!

The first showcase of the new version of DLSS was presented through Grace from Resident Evil 9: Requiem and she, as is already clear to everyone, looks terrible. Her face looks different and leans more toward the face of some model walking down the street, rather than a young, frightened woman preparing for the already well-known adventure. I personally believe that the skilled people from Nvidia did this on purpose to trigger negative reactions. Negative publicity is still publicity, and they have completely succeeded, because the entire internet is buzzing about it right now.

However, let’s just look at the other protagonist of RE9, our very dear Leon Kennedy, who, if you ask me at least, we must all agree looks undoubtedly significantly better. Unlike Grace, he doesn't lose any visual identity; he is still entirely himself. To be precise, the changes come down to how parts of his hair and face are lit, more emphasized facial hair on his chin, and slightly better-defined facial features.

It is necessary to immediately address the thing all critics agreed on, which is that this technology changes facial geometry, which is absolutely wrong. We must take two very important things into account: first, that we had the chance to see mostly static scenes without much action, and the second, much more important one – the change in facial geometry is an optical illusion caused by light, and it is present in real life as well. I know it sounds strange, but you can watch this video and see for yourself how your face could "change appearance" depending on the position of the light source.

So what's going on with those faces?

We need to linger a bit longer on character faces, which have undoubtedly sparked the most controversy. First of all, it's clear that in-game graphics have reached a point where we have photorealistic environments—just look at the water in the 10-year-old The Witcher 3, it looks great, or the environments in RDR2, which are still beautiful. But faces, well, that's a different story. Bringing faces to a level of real-time photorealism is practically one of the most "expensive" problems in computer graphics. The way skin reacts to light is not simple to replicate at all without making the character look lifeless or like a wax figure.

Let’s stick with Grace's face for a moment longer, but this time using a different image from the website where Nvidia provided some flawed explanations. However, what is definitely noticeable in this image is that she no longer looks like an AI-generated ad, but like a living person, and the model fully retains the original vision. Of course, I’m not here to hype something to the moon and sing nothing but praises, but to take a realistic look at the situation. Thus, we have to admit that not every game presented looked good. The application of DLSS to Hogwarts Legacy is, to put it mildly, strange. For me personally, the original style is completely ruined; the characters look too much like "AI slop." Something similar can be said for Starfield, which has rather poor face models; with DLSS, they definitely become more detailed, but not in a way that aligns with the style.

So, the problem that arises is the destruction of artistic intent.

Total control

Let’s get straight to the point—with what we know now, the answer is no, DLSS 5 will not destroy the developers' artistic intent. It should be clear to all of us that everything we read on the internet these days is based on a very small amount of information, but what has been confirmed by Nvidia is that DLSS 5 will come in the form of many independent tools, one of which will be masking. Masking is a form of control that involves giving the developer the ability to choose specific parts of the model where they want—or don't want—to apply DLSS. Likewise, if we set aside the obvious choice of not using the technology at all, the developer will be able to fully control the intensity of individual elements that make up DLSS 5. Therefore, the developer's control won't just come down to toggling it on and off, but to adjusting the intensity of specific parts—contrast, saturation, gamma, etc.

One user on X showed how that same model from Hogwarts would look if this technology were used at a lower intensity, specifically to only improve the lighting and the way it interacts with the model’s face. The result is definitely better, and the artistic intent is 100% preserved.

At the end of the day, developers can decide to use DLSS only on the environment, which could definitely raise the level of detail within the game. No matter how bad Hogwarts looked, if you look closely at the objects and certain parts of the architecture, you can definitely notice improvements. For now, from this limited showcase, it can be observed that almost all scenes are over-brightened, which creates a certain uncanny valley effect. Although we don’t know for sure, I assume that this will also be controllable and adjustable so that it looks a bit darker and better conveys the artist's original vision—because even if calculations say this is more realistic, it doesn't mean it's necessarily better.

What we as players must have a bit of understanding about is that this is the first version, just announced, and for now we have no idea what its true potential or current quality really is.

Hell no, it's far from perfect - it's actually worrying

In this text, I didn't dwell too much on the negative aspects, as the internet is already full of them and they are mostly all one can read about anyway. Some of them are a bit manipulative, as they don't show side-by-side images where it can clearly be seen that some are actually different frames, and in those where the frames are the same, only the change in how light interacts with the object is visible. What we should be talking about is the negative way this technology could impact an already monopolized market.

Nvidia has been a company fully oriented toward AI technology for some time now, and considering they control about 95% of the market, we can't help but worry about what is yet to come in the future. Establishing a monopoly, in the end, can never be good for us as users. Taking into account that U.S. President Donald Trump finds it "funny" that Nvidia controls almost the entire market, it's impossible to remain indifferent and not wonder what AMD's next moves should be.

The one last thing I would like to address and what has been confirmed to us is that two 5090 graphics cards were used for the demonstration of this technology, where one ran the games and the other was used specifically for DLSS 5. I'm not saying that this technology won't be expensive and that it won't be exclusive to likely the 50xx and newer series of Nvidia graphics cards, but what is certain is that you won't need two cards to run it. The reason why they did it this way is actually very simple – imagine that you are presenting something revolutionary and very controversial in itself to the whole world (and to a multitude of journalists who are ready to catch even the smallest details); the thing you want to be 100% sure of is that nothing goes wrong. That is also your answer to the initial question.

Nothing is black or white.

DLSS 5 represents, whether we like it or not, a fundamental shift in computer graphics. Brute-force calculations will be replaced—or rather, supplemented—by an AI that will "imagine" the light for us. While the early showcases have given us plenty of reasons to be skeptical, the technology itself isn't the villain—it’s just a new, incredibly complex brush in a developer's kit.

At the end of the day, one very important thing should be emphasized—DLSS 5, like all previous versions, is just a technology that a developer can choose whether or not to use. Some games will benefit from it, some won't, which is normal. This technology will be exclusive to Nvidia hardware, and it might not take off at all but instead become something like ray tracing—a toy for people with a lot of money.