Table of Contents
DLSS 5 Developer Controls Take the Spotlight at SIGGRAPH 2026
NVIDIA has shown a clearer look at how developers will be able to control DLSS 5, following months of debate around whether the technology risks making games look too processed. When DLSS 5 was first revealed, the pitch was ambitious. Rather than focusing only on upscaling or frame generation, NVIDIA described it as a new step in AI-assisted visual fidelity, using an AI model to enhance lighting, materials, reflections, skin, hair, fabric, and other scene details in real time.
That immediately raised a fair concern: who decides what the final image should look like? For players and developers worried about games being pushed towards the same glossy AI-enhanced look, NVIDIA’s answer has been artistic control. At SIGGRAPH 2026, the company finally showed more of what that control actually looks like.
DLSS 5 Will Include Multiple AI Models
One of the bigger takeaways is that DLSS 5 will not force developers to rely on a single model across an entire game. NVIDIA demonstrated multiple models, labelled Model A, Model B, and Model C, each producing a different visual result from the same scene. Developers will be able to pick the model that best suits a specific moment, rather than applying one look everywhere.
That means a studio could use one model for cutscenes, another for open-world traversal, and a different one for specific interiors or character-focused scenes. In theory, that gives art directors more room to decide when DLSS 5 should push realism further and when it should stay closer to the original render.
Structure and Tone Sliders Give Artists More Control
NVIDIA also showed 2 main intensity controls: structure intensity and tone intensity. Structure intensity affects higher-frequency visual detail, including ambient occlusion, reflections, contact shadows, and subsurface scattering. Tone intensity affects broader visual qualities, including lighting, colour, and the overall mood of a shot.
That separation matters because not every enhancement should be treated the same. A developer might want stronger contact shadows on objects without changing the mood of the scene, or richer lighting without adding too much texture sharpness.
The company also reiterated that DLSS 5 does not change the underlying geometry of the scene. The original renderer still controls the camera, composition, models, and world structure, while DLSS 5 works as an enhancement layer on top of the rendered frame.
Masking Could Be the Most Important Tool
The most interesting part may be selective masking. NVIDIA showed DLSS 5 being applied to specific objects, characters, or groups of props while leaving the rest of the scene untouched. Some masks can come from the model’s own understanding of the scene, while others can be authored through the engine by developers.
That means a studio could apply DLSS 5 to environmental materials, bottles, foliage, or reflective surfaces, but reduce or disable it on faces if the result looks too artificial. For developers worried about preserving character likenesses, scars, costumes, or carefully directed lighting, this kind of control could be essential.
NVIDIA Still Has to Convince Players
The SIGGRAPH demo makes DLSS 5 sound more flexible than its first reveal suggested. Instead of a one-size-fits-all AI pass, NVIDIA is presenting it as a controllable rendering tool that developers can tune by model, scene, object, structure, and tone. The company also says it is continuing to build more controls based on partner feedback.
Still, this does not end the debate. Players will ultimately judge DLSS 5 by how it looks in real games, not by controlled demos. If studios use it carefully, it could help scenes look richer without burying a game’s art direction. If used too aggressively, it could still produce the over-processed look that made some players sceptical in the first place.
For now, NVIDIA has at least made its argument clearer: DLSS 5 is not meant to replace the artist. It is meant to give developers another visual tool. Whether that promise holds up will depend on how much control studios actually use once the technology ships.
