OpenDLSS-NR · WebCuda

NEURAL RENDERING

Image workspace

NO INPUT

Comparison

INPUT—

↥Your next render starts here

Choose an image or 3D object to begin.

NR OUTPUT—

✧Ready when you are

Load a model, set your controls, and render.

Choose your NR DLL and an input to begin.

Native validation

Load a fixture folder to compare recorded network boundaries and output.

02 / RENDER

Control the network’s tone and structure response.

Local tone 1.00

MinimalFull

Structure 1.00

MinimalFull

Automatic structure maskSkin structureStyle conditioning

Network conditioning only; no additional colour grade is applied.

Noise seed

Temporal inputs

Optional previous output and motion for this frame. Files must match the selected resolution.

Previous outputMotion vectors

Float32 RGBA, little-endian. XY: current-to-previous UV offset. Z: validity. Without motion, history must already be aligned.

First frame · no history

Network input override

Precomputed features

Float32, 16 channels per padded pixel. Overrides image preprocessing and neural controls.

Reference renderer. Larger frames take more time and GPU memory.