ComfyUI Extension: ComfyUI-PIX_Nvidia_RTX_Deblur_Denoise

Authored by PixWizardry

Created

Updated

3 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

NVIDIA RTX Deblur & Denoise Nodes for ComfyUI\nThis extension provides GPU-accelerated deblur and denoise nodes powered by NVIDIA RTX technology via the nvvfx SDK.

Looking for a different extension?

Custom Nodes (0)

    README

    Nvidia_RTX_Deblur_Denoise_ComfyUI

    Two GPU-accelerated post-processing nodes for ComfyUI powered by NVIDIA RTX technology via the nvvfx SDK. Only works on NVIDIA RTX GPUs.

    This is a ComfyUI port of the Deblur and Denoise features from sd-forge-nvidia-vfx by Haoming02. Only these two features have been ported as a custom node pack for ComfyUI.

    Nodes

    RTX Deblur (image/postprocessing)

    Corrects blur and softness in images using NVIDIA's AI deblur model. Output is the same resolution as input.

    RTX Denoise (image/postprocessing)

    Removes noise and grain from images using NVIDIA's AI denoise model. Output is the same resolution as input.

    Quality Levels (both nodes)

    | Level | Description | |-------|-------------| | Light | Minimal correction, preserves most original texture | | Moderate | Balanced (default) | | Aggressive | Stronger correction | | Maximum | Strongest correction, may over-process |

    Requirements

    • NVIDIA RTX GPU
    • nvidia-vfx Python package

    Installation

    1. Place this folder in your ComfyUI custom_nodes/ directory
    2. Restart ComfyUI
    3. Search "RTX Deblur" or "RTX Denoise" in the node browser under image/postprocessing

    Example Workflows

    Basic deblur: LoadImage → RTX Deblur (Moderate) → SaveImage

    Basic denoise: LoadImage → RTX Denoise (Light) → SaveImage

    Combined: LoadImage → RTX Denoise (Moderate) → RTX Deblur (Moderate) → SaveImage

    Notes

    • Both nodes use zero-copy GPU memory sharing via DLPack — tensors stay on the GPU throughout
    • Batch processing is supported; batches are automatically split to fit within GPU memory limits
    • These are distinct from the RTXVideoSuperResolution node — they do not change image dimensions

    Credits

    • Original implementation by Haoming02 — sd-forge-nvidia-vfx
    • This project ports the Deblur and Denoise features from that extension into a standalone ComfyUI custom node pack

    Run ComfyUI workflows without the setup

    No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

    Learn more