Nvidia unveiled DLSS 5 in March, an AI-driven real-time upscaling and frame-generation system that can enhance not only resolution but also lighting, shadows, textures and even character faces in games. The announcement quickly became controversial after promotional imagery and leaked builds prompted criticism that the technology sometimes produces a ‘beauty filter’-like, artificial appearance.
What DLSS is and what’s new in version 5
DLSS (Deep Learning Super Sampling) traditionally renders at a lower native resolution and uses neural upscaling to make the output look as if it were running at a higher resolution (for example, 4K), reducing GPU load. Over the years the system has evolved — DLSS 4 could already predict future frames from prior frames.
DLSS 5 goes further: Nvidia says it inspects every frame in real time and refines appearance within the existing graphics engine, improving shadows, skin, hair and other details. The company also claims the approach is deterministic, so details remain stable between frames.
What triggered the backlash
One promotional image from March showed Grace Ashcroft from Resident Evil Requiem with a markedly smoothed, cosmetically enhanced face, which many viewers found unnerving. Jensen Huang, Nvidia’s CEO, pushed back on criticisms, but the debate escalated when DLSS 5 support was found in files for NBA 2K27 and leaked online. Users began applying DLSS 5 to many games and sharing videos that, in many cases, displayed what some described as overly filtered character faces and overly high contrast.
The two sides of the debate
- Proponents argue DLSS 5 delivers visible improvements in many titles: more realistic skin tones, richer clothing textures and better shadows. Daniel Vávra, co‑founder of Warhorse Studios, reported using DLSS 5 with Kingdom Come: Deliverance II and found it produced more lifelike textures and shadows without distorting facial features.
- Critics warn that heavy use of the technology could homogenize game visuals, particularly as many studios adopt engines like Unreal Engine 5. Early promotional materials and leaked videos reinforced fears that, if set too strong, DLSS 5 can erase individuality in character faces.
Performance and practical limits
Users testing DLSS 5 reported a significant performance hit — around a 50–60% drop in frame rate in some cases. Nvidia has responded that with all DLSS features enabled one can theoretically reach 370 fps in certain configurations, but that depends heavily on very high-end hardware; Nvidia’s example flagship card, the RTX 5090, was noted as costing over 2 million HUF in Hungary.
DLSS 5 is not a single on/off toggle: developers can choose among three base models, tweak intensity sliders, and define masks to exclude certain elements (for example, faces) from the effect. Many of the widely circulated videos likely used high intensity settings.
Where it helps and where it fails
DLSS 5 appears to perform well for games aiming at photorealism — NBA 2K27, the first officially supported title, was commonly cited as looking more realistic in skin, clothing and shadowing. By contrast, stylized or highly unique visual styles can be poorly served by a model trained toward photorealism, producing odd results.
Ethical and long-term considerations
The controversy also touches on broader concerns: large resource requirements, the risk that developers will increasingly rely on generative tools for convenience rather than artistic choice, and environmental and ethical questions tied to AI. For many, it is unsettling that AI could shape not only development workflows but also the final visuals players see.
Conclusion
DLSS 5 can be a meaningful technical advance for real-time rendering, delivering impressive results in some games and settings. Equally, improper or excessive use can yield artificial-looking results and risks visual homogenization. The ultimate impact will depend on developer choices, sensible configuration of the feature, and how the community judges deployments in real titles.
