DLSS 5 Details Revealed NVIDIA Confirms It Only Uses 2D Frames

DLSS 5 has remained controversial since its reveal, which prompted the NVIDIA CEO to address the criticism. NVIDIA has now stepped in with a clearer explanation of how the technology works. While the company aimed to address confusion, the details appear to reinforce many of the concerns raised by gamers.

DLSS 5 relies only on a 2D rendered frame

According to information shared via VideoCardz, DLSS 5 works by using a single 2D rendered frame combined with motion vectors. Unlike previous expectations, the model does not access deeper data from the game engine, such as geometry, depth buffers, or material maps.

NVIDIA confirmed that the system does not read metallic, roughness, or normal map data. Instead, it derives all visual understanding directly from the final rendered image, which represents a major shift from traditional rendering pipelines.

The AI infers details instead of reading real data

DLSS 5 analyzes the image and attempts to reconstruct scene elements through inference. It can recognize characters, hair, skin, and fabric, as well as different lighting conditions such as front-lit or back-lit scenes.

However, because the model relies entirely on interpretation rather than actual scene data, it effectively predicts how elements should appear rather than reproducing them exactly as designed by developers.

Concerns grow over altered visuals

This approach has raised new concerns, especially after early preview footage surfaced. Some clips appear to show noticeable changes in facial features, hair detail, and even subtle “makeup-like” smoothing effects.

NVIDIA maintains that the underlying geometry remains unchanged, but the final output can still look different. For many players, this creates uncertainty around whether DLSS 5 preserves the original artistic intent of a game.

Developers have limited control over AI output

NVIDIA also clarified that developer control remains relatively limited. Developers can adjust parameters such as contrast, saturation, gamma, and blending, or apply masking to disable effects in certain areas.

However, they cannot directly correct specific AI-generated changes. If the system alters a face or adds unintended detail, developers must rely on indirect methods like reducing intensity or masking the feature entirely.

DLSS 5 operates only in screen space

Another important detail is that DLSS 5 works entirely in screen space. It has no awareness beyond the current frame and does not maintain a deeper understanding of the scene across frames.

This means the AI generates its output based only on what it sees in a single image, without access to the full context of the game world.

NVIDIA defends the vision, but backlash continues

NVIDIA positions DLSS 5 as a tool designed to enhance artistic vision using generative AI. The company highlights its ability to improve perceived detail and visual quality.

However, the community reaction remains mixed, with ongoing concerns about AI altering original art direction and the lack of precise control. There is also no confirmation yet on whether DLSS 5 will support hardware-agnostic standards like DirectML.

A major shift with uncertain reception

DLSS 5 represents a significant evolution in rendering technology, moving beyond upscaling into full AI-driven image generation. While the approach opens new possibilities, it also introduces questions about accuracy, consistency, and creative control.

NVIDIA’s clarification answers how DLSS 5 works, but it does little to ease the backlash that has followed the technology since its announcement.