Nvidia’s so called DLSS 5 neural rendering pipeline is out in the wild in an extremely unfinished, unofficial form, and it is already producing some wild results. I have seen the opinions flying around, so I wanted to look at what it actually does to real games.
There is one enormous caveat before anybody gets too excited or too terrified: this is not a finished version of DLSS 5. It appears to have originated from a pre-release build connected to NBA 2K27, and modders have adapted it to games it was never designed to support. That means disabled safeguards, crashes, visual glitches, rough performance, and all the other stuff that makes it a terrible daily-driver mod.
I would not recommend installing it unless you are technically confident, willing to troubleshoot a mess, and very comfortable with the possibility that it simply refuses to work. The sensible move is to wait for final game-specific implementations from developers. Still, as a glimpse of where neural rendering may go, it is pretty interesting.
One important point: DLSS 5 does not add geometry. A low-poly object remains low-poly. What it can change, sometimes dramatically, is the apparent detail in textures, lighting, cloth, surfaces, shadows, and faces. The trouble is that faces and skin are also where this early implementation tends to go completely overboard.
Key Takeaways
- This unofficial DLSS 5 build is unstable, crash-prone, and not recommended for ordinary game installs.
- Control and The Last of Us Part I show the clearest visual potential, especially in lighting-rich scenes.
- Skin and faces need much subtler processing than cloth, foliage, buildings, and other environmental detail.
- DLSS 5 cannot replace a proper remaster because it does not add new geometry, models, or artist-led upgrades.
Table of Contents
- 5. Control: The Best Case for Subtle Enhancement
- 4. Watch Dogs: Daylight Turns Into Goo
- 3. Grand Theft Auto 4: Great Environments, Overcooked Faces
- 2. The Last of Us Part I: Moody Lighting Helps a Lot
- 1. The Witcher 3: Better Lighting Is Not Enough
- What DLSS 5 Needs Before It Is Ready
5. Control: The Best Case for Subtle Enhancement
Control is probably the cleanest example of what I would want this technology to do. It does not bulldoze the game’s visual identity and replace it with something unrecognisable. It still looks like Control, just with a little more definition in the right places.
Character features gain some extra depth, particularly around skin, eyes, hair shadows, and the way a nose catches light. It is not a radical model upgrade, because it cannot be, but it can make those existing details feel more present. The game’s moody concrete interiors and sharp pools of light also suit the effect really well.
Even here, though, the stability situation was rough. This was one of the better-running tests, and it still could go haywire after about four minutes. I ran into an issue where the DLSS 5 output appeared at a much lower resolution than the game and was laid over the full-resolution image. It looked as weird as that sounds. Crashes were common too.
That does not mean the underlying idea is bad. It means this hacked-in build is nowhere near ready. In a proper release, the useful result would be a modest pass that enriches lighting and material detail while respecting the artists’ original image.
4. Watch Dogs: Daylight Turns Into Goo
Watch Dogs is where the experiment gets much uglier. In daytime scenes, this looked rough as hell. The whole image seemed compromised, and Aiden Pearce in particular could take on a distinctly goo-like look.
Part of that may have been the same low-resolution overlay issue I experienced elsewhere, but whatever was happening internally, the visual result was not doing the game any favours. Watch Dogs was always a game people wanted to see running at the level originally promised, but this is not that. It is not full glory. It is a broken filter sitting on top of a game that already looks decent.
Night scenes and cutscenes held together better. Under darkness, streetlights, and artificial lighting, the flaws were easier to hide and the effect could add some environmental richness. Performance was also less immediately disastrous in this game, at least in some test sessions, but that was not remotely consistent enough to call it viable.
The useful lesson from Watch Dogs is about control. I would want developers to be able to mask character skin and reduce the intensity of the effect on faces, while allowing a stronger treatment on clothing, buildings, weather, streets, and other environmental elements. When it works, those background surfaces are often the interesting part. The human faces are where the tech starts looking deeply strange.
3. Grand Theft Auto 4: Great Environments, Overcooked Faces
Grand Theft Auto 4 can look genuinely impressive with this setup, but it comes with a big asterisk. This is not vanilla GTA 4 suddenly transformed by DLSS 5 alone. The test footage also uses RTX Remix, which is needed to bring DLSS compatibility into the game and already adds higher-resolution textures and new RTX lighting features.
So, credit where it is due, but credit has to be distributed properly. The result is a mixture of multiple technical changes rather than a simple before-and-after DLSS 5 demonstration.
Even with that qualification, the potential is easy to see. Cloth can look amazing. Street surfaces, lighting, and the inanimate stuff in a scene gain a degree of richness that feels like a meaningful upgrade. But skin still gets pushed too hard.
The closest comparison I can make is old-fashioned oversharpening. An overly sharpened image gets this artificial texture that makes everything feel harsh and wrong. DLSS 5 does not always go that far, but on faces it can create the same general feeling. Some skin enhancement is good. Too much creates exactly the AI-generated slop effect people hate.
The frustrating part is that a setting strong enough to make a character’s face look acceptable can also weaken the benefits on the environment. This is why a single global strength setting is not enough. The final technology needs different treatment for different parts of the image:
- Lower intensity for skin and faces to preserve natural facial detail.
- Higher intensity for cloth, where the added texture can look fantastic.
- More latitude for backgrounds, rigid objects, weather, and distant surfaces.
- Developer-level masking tools rather than a universal maxed-out player toggle.
That kind of restraint is the difference between an enhancement and a visual wrecking ball.
2. The Last of Us Part I: Moody Lighting Helps a Lot
The Last of Us Part I has some of the most impressive moments of the whole experiment. Its naturally moody, cinematic lighting gives this neural rendering effect a lot to work with.
The game frequently puts characters in dark spaces cut by a single window or harsh directional light. Those shadows carve across faces, silhouettes stand out, and the material response of skin is a huge part of the look. DLSS 5 seems to add a little extra grit and cinematic richness in these scenes. I cannot definitively say it is creating proper subsurface scattering in skin, but the way light appears to interact with faces can look more convincing at times.
Outdoor scenes generally came out better than indoor ones. The treatment of foliage, brickwork, ruined buildings, and natural light was particularly effective. In really dark indoor spaces with no meaningful light source, however, it could look much less convincing.
And yes, I am still coming back to skin. The same problem remains: the effect needs to be gentler on people. The Last of Us Part I benefits from this more than most because its lighting already has such strong mood and contrast, but it still demonstrates why developers need granular control.
1. The Witcher 3: Better Lighting Is Not Enough
The Witcher 3 was the least convincing use case of everything I tested. There are individual shots where the lighting looks really cool. Certain skin interactions with light seem a bit improved. But the experience was flickery, temporally inconsistent, and just not good enough to justify the effort.
That matters because there is a common claim that technology like this makes traditional remasters obsolete. Just make an old game compatible with a modern operating system, throw neural rendering at it, and suddenly you have a remaster. I do not buy that at all.
A real remaster can improve models, textures, lighting, effects, systems, and art direction in a coordinated way. This early DLSS 5 implementation cannot do that. It cannot add geometry, and it cannot substitute for proper artist-led work. In the case of The Witcher 3, the official remaster upgrade is a much more substantial proposition, adding its own graphical features without requiring leaked files, brittle mods, removed safeguards, or a PC that feels like it is trying to melt itself.
Would it be cool to get better lighting, ray tracing, and more convincing skin shading in The Witcher 3? Obviously. But this is not a shortcut to that future. For now, it is an unstable proof of concept that does not make much sense as a way to play the game.
What DLSS 5 Needs Before It Is Ready
I am not writing off DLSS 5. There is real utility here. Control, parts of GTA 4, and the outdoor scenes in The Last of Us Part I show that neural rendering could become an interesting artistic tool.
But the final version needs artists and developers with restraint. It needs scene-aware masking that can identify skin and apply a subtle effect there, while allowing stronger detail enhancement across clothing, architecture, foliage, weather, and other inanimate materials. It also needs stable, game-specific integration rather than unofficial files that have been forced into games with their limits stripped out.
For broader context on the technology family, Nvidia provides an official overview of DLSS. The interesting question is not whether a hacked pre-release build can make an old game look bizarrely different. It absolutely can. The question is whether developers can use a finished version with enough precision to make games look better without making people look like they have been aggressively run through an AI filter.
Right now, this is not releasable as a general-purpose product. It is finicky, unstable, and often ugly. But there are enough compelling moments that I am interested to see what a final, properly integrated release can do.
For more gaming breakdowns and strange experiments like this, you can subscribe to gameranx.
