DLSS 5 Is Changing How Your Games Look—Here’s How the AI Works

With DLSS 5, NVIDIA uses generative AI to change how light and materials look in games. That includes the way light filters through leaves, how skin responds to light and how fabric shines. The aim is a more lifelike image—but the technology also demands processing power and could alter a game’s visual style.

DLSS stands for Deep Learning Super Sampling. The name comes from the world of image upscaling, but NVIDIA uses it to describe several different technologies. Some help graphics cards render images more efficiently. Others change image quality or generate additional frames.

With DLSS 5, NVIDIA is introducing 3D-Guided Neural Rendering: an AI model that takes an existing game image and generates additional lighting and material detail. That sets it apart from other features in NVIDIA’s DLSS technology.

Earlier DLSS features also use AI and generate image information that was not rendered directly. The difference is what they do: DLSS 5 also draws on learned knowledge of the real world to change the final image.

The game engine—the software that drives and renders the game world—still provides the foundation. It calculates where objects are, how characters move and what the scene looks like. According to NVIDIA’s explanation of neural rendering, DLSS 5 then processes the rendered image in an additional step.

Imagine a character standing by a window. The game determines the shape of their face, their clothes and their position relative to the window. The AI model can then adjust how light falls on their skin and how their clothes shine. It doesn’t create a new 3D character; it changes how the existing character appears on screen.

In its technical overview of DLSS 5, NVIDIA says the model was trained on visual data from the real world. It uses a specialised diffusion model, a type of generative AI that creates images based on learned patterns. DLSS 5 uses this model to produce an adjusted game image in a single generation step.

The model uses the current game image, motion vectors, information from previous frames and developer settings. Motion vectors describe how elements in an image move. They help keep added detail attached to the same surface as the camera turns.

After all, frames need to look convincing individually and fit together over time. A reflection that jumps around whenever the camera moves can shatter the illusion. That’s why NVIDIA designed the model to produce predictable results and maintain consistency between consecutive frames. That design goal doesn’t rule out visual glitches.

In local gameplay, the supported graphics card handles this AI step. Players don’t need to write text prompts, and each frame isn’t sent to a separate online image generator for processing.

This is a specific use of AI in games: the model adjusts the visuals as you play. It doesn’t write new missions or control enemies. AI characters you can talk to rely on different systems, too.

A game doesn’t have to look like live-action footage to look good. Creators may deliberately choose flat colours, exaggerated faces or unrealistic lighting. In that context, extra skin detail or softer lighting could undermine the intended style.

Even an unchanged 3D shape can look different under new shadows, highlights or skin tones. Preserving the geometry, then, doesn’t mean preserving every artistic choice. The key question isn’t just how much detail AI adds—it’s whether that detail fits the game.

NVIDIA gives developers artistic control settings to manage the effect. They can choose between different models, adjust the strength of detail and lighting, and use masks to decide which parts of the image the AI can change.

The technology does not automatically replace ray tracing or path tracing, either. Those techniques calculate how light moves through a virtual scene. Neural rendering builds on them using learned visual patterns. Lighting that looks plausible because an AI model generated it is not the same as a separately calculated path of light.

DLSS 5 does not guarantee a higher frame rate. Neural rendering adds an AI process to the existing rendering workload, which can actually lower frame rates. The impact depends on factors such as the graphics card, resolution, model and how the game implements the feature.

That differs from Super Resolution, where the game renders at a lower resolution and AI reconstructs the image at the output resolution. The savings from rendering fewer pixels can outweigh the cost of reconstruction. Neural rendering, by contrast, adds another step to change the image’s appearance.

The features can work together, but turning several settings on at once does not show how much each one contributes. Comparing all DLSS options enabled with all of them disabled is therefore not a clean test of neural rendering.

Frame Generation calls for another distinction. It can make motion look smoother by adding intermediate frames, but the game does not process new input proportionally more often. That is why more FPS thanks to AI does not automatically mean a more responsive game. Look beyond the FPS counter: check the frame rate before frame generation and the delay between moving your mouse and seeing the result.

Both the game and the hardware must also support the feature you want to use. A graphics card that supports Super Resolution may not support every other DLSS feature. Check the official DLSS compatibility list and the game developer’s information.

Compare neural rendering on and off in the same scene, keeping all other graphics settings unchanged. Move the camera, inspect fine details and watch for changes to faces, shadows and reflections. A convincing still image is only a real improvement if the visuals remain stable during gameplay and the controls still feel responsive.