DLSS 5 on your Blender renders

What happens to your render

DLSS 5 Neural Rendering does not upscale and does not denoise: it reworks lighting and materials. NVIDIA describes it as infusing the scene with photoreal lighting and materials, and gives developers controls for intensity, colour grading and masking. On a Blender render that means skin, fabric, metal and shadows reinterpreted by the model, with the geometry left where it was.

How far it goes is up to you and the parameters, which is why the preview exists: a stylised or deliberately flat render can come out transformed in a way you did not want, and seeing it on one frame before exporting a whole animation saves you finding out at the end.

What a render file cannot give the model

In a game, DLSS 5 takes each frame's colour and motion vectors from the engine, and NVIDIA says it is trained to recognise engine data — surface albedo, lighting, normals. A PNG or an MP4 contains only pixels.

For a clip, the motion between one frame and the next is estimated from the images themselves, with optical flow; for a still image there is no motion to estimate. Your scene's depth, normals and masks do not go in: Blender can render those passes, but the Studio does not take them and works from the pixels.

Which is how to read the result: a faithful preview of what the model does to that image, not a prediction of how the same scene would look in a game that integrates DLSS 5. For comparing styles, looks and intensities it is exactly what you need; for judging an integration it is not.

The DLSS coming in Blender 5.3 is something else

Blender 5.3, in alpha until 30 September 2026, adds a DLSS viewport denoiser to Cycles on NVIDIA GPUs: it makes the viewport cleaner and more responsive by upscaling and reusing samples from previous frames. NVIDIA presents it as DLSS 4.5 Ray Reconstruction.

It is not DLSS 5. It is a denoiser, it works in the viewport rather than on the final render, and it uses the scene's passes — colour, depth, albedo, normals, roughness, motion vectors. DLSS 5 does the opposite: it starts from a finished image and changes its light. People who search «DLSS 5 Blender» often find the first and wanted the second.

The community addons

As of September 2026 there are two projects on GitHub that bring DLSS 5 into Blender. Both state that they are unofficial and not affiliated with NVIDIA or the Blender Foundation, and both need an RTX 40 or 50 on Windows plus the DLSS 5 runtime, which they do not include.

And both work from pixels, like the Studio: one captures the Blender window and estimates motion with optical flow, the other processes still images only. The difference is not in the result but in where it runs: on your card, with a file to find yourself and a licensing question that stays yours, or on ours, by uploading a file.

What happens to the file

It stays on the disk of the machine holding your slot, for the length of the slot, and is deleted when the slot expires. It does not go through a cloud and is not copied anywhere else.

Publishing it to the catalogue is your choice, one frame at a time: if you do not publish it, nobody sees it.