Cyberpunk provides an immediate demonstration of why DLSS 5 is difficult to judge as a straightforward graphical upgrade. Turning it on can deepen shadows, alter the apparent lighting around characters, and make clothing or facial details stand out dramatically, yet those changes do not always preserve the original image. The hosts are appropriately cautious about declaring those results inherently better, repeatedly distinguishing between added realism and a game actually looking the way its artists intended.
That caution matters because the build being demonstrated is not an officially supported implementation across these games. The DLSS 5 DLL has reportedly been taken from NBA 2K27 and modified to operate elsewhere, giving the group access to developer-facing controls such as three models, intensity, tone, and structure adjustments. The presentation consistently reminds viewers that Nvidia may tune individual games differently, which makes the demonstrations useful as an early exploration of the technology rather than a definitive review of its finished form.
The developer controls produce some of the most informative comparisons. Changes to tone and structure can make a shirt emblem clearer, deepen apparent material detail, alter skin shading, and make incidental Cyberpunk NPCs look much closer to major characters. At the same time, the technology occasionally appears to invent or exaggerate information: one character suddenly looks more muscular and defined, while faces can acquire wrinkles, pores, veins, makeup-like shading, or altered age characteristics that were far less apparent in the original rendering. The hosts' joking reactions are entertaining, but they also illustrate the central technical concern—greater perceived detail is not necessarily greater fidelity to the source.
Final Fantasy VII Rebirth demonstrates that problem particularly well. Fraying fabric, armor scratches, rocks, and environmental objects can gain impressive definition, but stylized faces are pulled toward a more photorealistic appearance that clashes with the game's established aesthetic. The resulting "realism filter" effect on Cloud and other characters is a persuasive counterexample to the idea that additional AI-generated detail should automatically be considered an improvement. Similar concerns emerge elsewhere when dreamlike color grading and stylized lighting are replaced by heavier shadows or more naturalistic tones.
Other examples show why dismissing the technology outright would also be premature. Close-up faces can gain strikingly convincing skin texture, lips, eyes, hair definition, and small imperfections, while certain landscapes and background characters benefit noticeably from enhanced structure. The presentation is strongest when the hosts keep changing their minds from scene to scene, because that inconsistency appears to be the real finding: DLSS 5 can look excellent in one view and immediately uncanny in another. Testing movement rather than relying only on carefully framed still scenes is also a useful priority, although crashes and severe hitching limit how extensively that question can be explored here.
Performance is the largest practical obstacle shown. The hosts report roughly a 50 to 60 percent performance cost, with even their RTX 5090 system struggling in places, and one comparison drops into extremely low frame rates. They speculate that mainstream hardware capable of comfortably using the feature may be years away, but that forecast is their assessment rather than something established by these demonstrations. Combined with the unofficial implementation, occasional crashes, and uneven visual results, the evidence supports excitement about DLSS 5's potential far more strongly than it supports treating the technology as a compelling reason to purchase new hardware today.
Pros
- Uses multiple games and varied visual styles to expose both impressive enhancements and obvious failures rather than presenting only favorable examples.
- Developer-facing tone, structure, intensity, and model controls provide unusually informative insight into how strongly DLSS 5 can reshape an image.
- Clearly acknowledges that these are unofficial implementations and repeatedly avoids treating the results as representative of final developer-tuned releases.
- The comparisons identify a meaningful distinction between increased photorealism, perceived detail, and preservation of a game's intended artistic style.
- Performance observations on an RTX 5090 make the substantial computational cost impossible to overlook.
Cons
- The unofficial, transplanted DLSS 5 implementation means visual defects, crashes, and performance measurements cannot establish how properly supported versions will behave.
- Several comparisons rely heavily on subjective visual reactions rather than a more systematic examination of temporal stability, motion artifacts, or repeatable image-quality criteria.
- Still-image pixel peeping dominates much of the testing despite the hosts themselves identifying real gameplay and movement as crucial unanswered questions.
- The extended sponsor segments and frequent joking occasionally interrupt an otherwise valuable technical exploration.
DLSS 5 emerges as a fascinating but deeply inconsistent attempt to enhance rendered imagery, capable of producing remarkable material and facial detail while also rewriting artistic choices in ways that can feel uncanny or simply unnecessary. The hosts do a good job resisting the temptation to label every increase in realism an improvement, and their unofficial testing exposes exactly why developer control will matter. As an early look at potential rather than a verdict on finished technology, it is revealing, entertaining, and appropriately skeptical.
