sNeuralRender

NVIDIA DLSS 5 Neural Rendering now runs inside Nuke.

sNeuralRender is a native Nuke plugin that drives NVIDIA's Neural Rendering runtime (the neural "relighting / detail" pass behind DLSS 5) directly in the node graph. No game, no capture, no ReShade: connect your footage, get the neural result back as an image.

First submitted: 13 October 2026

Author: Aleš Ushakou

Compatible Nuke versions: 15.1 or later

Compatibility: Windows


**What it does:**

• The same controls NVIDIA shows in its developer panel: global structure and tone, the

model's automask with skin structure, three models.

• A mask input in the usual Nuke sense: a channel that limits where the neural result lands.

• DLAA stage and DLSS upscale (1x to 4x) in front of the neural pass. Upscale first, then

neural render at the output size: visibly more detail than either alone.

• Highlight recovery: scene values above white keep their level and colour.

• Temporal history for stable sequences, with motion from NVIDIA Optical Flow or your own

SmartVector / VectorGenerator vectors, cut detection, replay when you scrub.

• About 15 ms per 1080p frame and 33 ms per 4K frame on an RTX PRO 6000. Tested on GeForce

RTX 5090 as well.

• Nuke 15.1, 16.1, 17.0 and 17.1

The NVIDIA runtime itself is not included (it is NVIDIA's software). A one-click script in the

download fetches it from public mirrors and verifies it; the README covers the rest.

Before / after, five pairs, drag the handle yourself:

https://alesushakou.github.io/sNeuralRender/

or register to download or rate.

Comments

or register to join the conversation.

Download Verification