Runs NVIDIA's DLSS 5 Neural Rendering model over your generations as a post-process. Adds material and lighting detail in ~10 ms (about 60 ms at 2 MP). No LoRA, no extra sampling steps, no structure changes — hands and faces stay exactly where they were. Works with any base model. Ill and Anima in versions - to get more reach. Download latest version.
What it actually does. The "NR" is Neural Rendering re-renders material and lighting response from colour, so it adds detail rather than removing it.
Requirements: Windows x64 + NVIDIA RTX (realistically 50 series), Forge on Gradio 4.40, recent driver.
Presets: Balanced (default), Full, Subtle — measured from a 54-setting × 15-image sweep.
Known limits: slight banding on large smooth gradients; glossy speculars compress; flat cel art loses 10–20% line depth at higher settings; the Skin structure slider does nothing (documented rather than hidden).
Installation:
Can be found in the dropdowns below in txt2img, Extras etc.
Unzip into
<forge>/extensions/Restart Forge
Credits: bridge and original ComfyUI node by codon2001, originally shared by 叁贰玖零 (3290).
Description
Works with every basemodel and photos.
Fixed a bug with dark/black parts becoming blue.
New Accumulate frames slider (1–8, default 1) for banding-heavy images