Nvidia snuck me into a room to show off DLSS 5 in person, and it's somehow beautiful, ugly, impressive, and troubling all at the same time

I think it's fair to say that DLSS 5 is the most controversial Nvidia creation… well, ever. After the huge amounts of backlash it received in reaction to its unveiling earlier this year, Nvidia's AI image enhancing tech has been mostly kept under lock and key.

However, myself and a select group of journalists were ushered into a small room in the Nvidia booth at Gamescom last week, amid some furtive glances and hushed whispering. As the lights went down, it became obvious as to why. Here was the latest version of DLSS 5 running in front of us, the controversial graphics tech in the flesh. And the message Nvidia wanted to get across was deliberate, and strongly put: It's all about developer control.

I was shown a character sitting within the Nvidia Zorah demo, the same shown at SIGGRAPH 2026 earlier this year. As he sat, looking around at his environment in a bored fashion, I leaned forward in my chair. DLSS 5 was obviously altering the image to a significant degree, but seeing the results in person, it was clear this was nowhere near the "yassified Grace" level of AI enhancement shown in the first reveal.

Latest Videos FromPC Gamer

Our digital example, casually inspecting the back of his hand, looked a lot like the original character model—but with more photorealistic lighting, and plenty of finer details that seemed fitting for the scene. It was nowhere near as brash an image as I was expecting, I'll say that much for free.

At the side of the screen was the DLSS 5 developer control menu. Two sliders, marked "structure intensity" and "tone intensity" were set somewhere around the midway mark—and at these settings, the character looked pretty damn good.

There are more tools for developers to play with, too. The masking feature allows devs to select individual items in a scene (say, a fruit bowl on a table) and give it the AI-enhanced treatment, while leaving everything else DLSS 5-free. The varying AI models can also be switched in and out (or on and off) on a whim.

According to Nvidia, the fine-tuning abilities of DLSS 5 are designed to be as developer-friendly as possible, allowing them to dip in and out of its features on a model-per-model basis to suit their artistic intent. Seeing the demo running in front of me with my own eyes, I could feel myself coming around to the idea. My arms uncrossed.

And then my hosts began messing with the sliders and the model presets, and an essential truth emerged. While a light dusting of DLSS 5 can cause a substantial visual leap in terms of photorealism, turning it up to 11 can still provoke an unwanted, gut-level reaction.

Turning DLSS 5 up to 11 can still provoke an unwanted, gut-level reaction

There are three image-transforming AI models for devs to choose from (at least in the demo I was shown): A, B, and C. While models A and C looked pretty good to me at various settings, model B made me wince when both sliders were pushed up. The HDR-like shadows became overblown, the skin textures too shiny, the facial features too keenly drawn.

To my eyes, characters with model B turned up can look substantially different from the model underneath, in a way that's reminiscent of the AI beauty filters we've all become far too familiar with. In short, it's still capable of giving me the ick.

I'll call it as I see it: DLSS 5 turned up to the guns, particularly using model B, can look downright ugly. But Nvidia's point makes a whole lot of sense. It's the developers, the artists behind the game, that choose how far to push the technology. If it oversteps the mark? Devs can adjust the settings and rein it back in, and you can turn it off entirely (separate from regular DLSS upscaling) if you disagree with their choices.

On that note, some Nvidia staff members I spoke to took issue with the idea that DLSS 5, even in the experimental version shown earlier this year, is capable of transforming the facial geometry of a character model. I argued the contrary, because well, look at the screenshots below.

Apparently, this is more of a perceived effect created by the lighting than an actual, DLSS 5-induced facelift. This eventually turned into a philosophical debate, where I argued that, even if it simply looks like the model is creating markedly different facial geometry from the original character models, the end result is functionally the same.

Because that's what it looks like, and what it looks like is the whole point. This argument resulted in a stalemate, so I'll let you be the judge for yourselves.

Anyway, the longer the demos continued, the clearer it became that DLSS 5 is capable of creating some stunning imagery when handled with a delicate touch, and some seriously overblown results when pushed to the limit. DLSS 5 can enhance—but it can also alter an image with all the subtlety of a sledgehammer.

(Image credit: Nvidia)

The example above shows a DLSS 5-enhanced scene of some foliage against a building, with ray tracing turned on and off. It's supposed to demonstrate that the more lighting information you give the models to work with, the better the visual result. The DLSS 5 and ray-traced image combined, though, looks seriously oversaturated to my eyes. Realism has gone out of the window, and the end result is… lurid, to say the least.

What is beautiful to some is garish to others, but all I'll say is that the term "better" is far from my mind. There's no accounting for taste, is there?

(Image credit: Nvidia)

What we're going to be relying on for the most part, though, are the tastes of the game developers themselves. I was taken into a separate room to try out NBA 2K27, the first game you'll be able to play with DLSS 5 enabled. Officially, at least, as it can apparently be modded into any game.

This was a very locked down session, in which the devs refused to be drawn in on the details of exactly what sort of DLSS 5 implementation I was looking at. However, I'd guess that, by Nvidia's standards, everything I saw was achieved by using the lower end of the DLSS 5 spectrum.

If developers tune it wisely, there are real visual benefits to be seen

The sweat running down the players' arms looked… sweatier. More glistening. The facial features looked more realistic, with better lighting in the nooks and crannies of their various mugs, and less of the "uncanny valley" effect I'd seen at higher settings. The crowd in the background transformed from low-poly models into a more passable depiction of basketball match attendees, without straying too far from the look of the original assets.

To my surprise, I didn't experience the AI image ick. It looked more like the difference between a game running on a console, and one running on PC with a ton of extra graphics options added on top. DLSS 5 doesn't look quite so revolutionary at these settings, perhaps—but the results impressed me more than I expected.

I was careful to look for signs of ghosting, or any morphing AI faces under fast motion. One of my critiques of the previous DLSS 5 footage was the temporal handling, which looked a little blurred and smeared to my eyes. However, there was no hint of that here. I played virtual basketball, I sucked at it, and I enjoyed the look of the game overall.

In all honesty, it felt like playing a next-gen game with cutting-edge graphics technology, and not an ill-advised AI gaming revolution. This, beyond anything else, was perhaps the best demonstration that DLSS 5 can be used to genuinely enhance the look of our games, rather than turning them into an AI-slathered mess. If developers tune it wisely, there are real visual benefits to be seen.

All of this might be a moot point, though, if performance is a concern. The Nvidiadevs I spoke to estimated a 50-60% performance hit with DLSS 5 turned on, which doesn't vary on a game-by-game basis. Essentially, it's still a massive resource hog as things stand, although it no longer requires two RTX 5090s just to run. Goodie.

Further performance improvements are forthcoming, says Nvidia, but my eyebrows jumped so high at this news I'm surprised they didn't shoot off the top of my head. Graphics card prices are outrageous right now, and I can't help wondering who's going to want to slice their frame rate in half in return for some AI-enhanced imagery, good-looking or otherwise.

All of the performance charts I was shown used a combination of 6x Multi Frame Generation and various levels of upscaling to bump up the numbers. And while I'm not as much of a frame generation sceptic as some, I doubt this will come as good news to those curious to try out the tech.

If the performance impact was minimal, I could see some gamers experimenting with DLSS 5 for themselves and potentially changing their minds. With this sort of massive dip, though, I think DLSS 5's GPU-hogging ways are a much bigger issue for adoption than the AI-ified nature of the tech itself.

(Image credit: Nvidia)

What's left, then, is the thorny issue of the ethical debate. The elephant in the room, and there's no getting around it, is that generative AI usage in games is a massive hot topic at the moment. In fact, AI usage in general is causing all sorts of debates in all sorts of arenas, from the potential effects on the job market, to ecological concerns, to discussions around the very nature of truth.

We're going to have to decide for ourselves if we want to engage with it

And this, I feel, is the real hurdle Nvidia is trying to jump. While it's clear that DLSS 5 can be beneficial to the image quality of our games in the right hands, whether we, as gamers, want to participate in the AI-ification of everything in this way is much more doubtful. And it's on this level where I feel far more conflicted about the tech.

My criticisms of the AI revolution, in a general sense, are well documented. And while I can absolutely see the potential benefits of DLSS 5, I left the demo booth with mixed feelings. Is this where the games industry is heading? Will I be spending most of my gaming time looking at more AI-enhanced imagery in the future?

(Image credit: Nvidia)

Will developers make use of the tech wisely, or will major releases eventually go for the full yassified approach, transforming our games into something unrecognisable from what they are today? Will devs jump in with both feet, or will they be more cautious? And crucially, is AI-enhanced imagery really what we want, even if it's capable of looking nice?

In principle, the demos I was shown should have me jumping up and down with excitement at the potential of better gaming graphics. But the AI association, perhaps even infiltration, into a form of media I truly adore is enough to keep me awake at night. In practice, DLSS 5 is certainly capable of creating some good-looking imagery in the right hands.

But it also represents a major leap towards an AI-enhanced gaming future, and one that still makes me, in the depths of my bones, more than a little uncomfortable. Whatever the case, DLSS 5 will be launching on September 3—and we're going to have to decide for ourselves if we want to engage with it.

Best graphics card 2026