Image & Video

New ComfyUI node brings 2-pass Nvidia RTX upscaling, rivals Topaz AI

Upscale AI videos with DeBlur then VSR in seconds, even on 4GB VRAM.

Deep Dive

Developer Deno released a new custom node for ComfyUI called Deno RTX Video Super Resolution, bringing Nvidia’s hardware-accelerated upscaling to AI video workflows. The node exposes all four Nvidia RTX upscaling modes — DeBlur, DeNoise, High Bitrate, and Video Super Resolution (VSR) — and introduces a novel 2-pass method. The first pass applies DeBlur to sharpen blurry AI-generated footage, followed by a second pass using High Bitrate or VSR for final upscaling. In tests, this sequential approach produced noticeably better results than single-pass VSR, effectively eliminating the need for paid tools like Topaz AI.

Performance is impressive: on an RTX 5090, upscaling from 512x512 to 1024x1024 takes about 5 seconds. For users with limited hardware, a batch image workflow requires only 10GB RAM and 4GB VRAM, completing the same upscale in 1-2 minutes. Installation is streamlined via ComfyUI Manager — search for "deno-custom-nodes" — with an included install_rtx_vfx helper script that automates backend setup, avoiding common errors. For early adopters, manual installation from GitHub is also available. The update is currently rolling out to the ComfyUI registry.

Key Points
  • Four Nvidia RTX upscaling modes (DeBlur, DeNoise, High Bitrate, VSR) with a novel 2-pass approach.
  • RTX 5090 completes 2x upscale (512→1024) in ~5 seconds; batch mode works on 4GB VRAM in 1-2 minutes.
  • One-click backend installer eliminates common setup errors; available via ComfyUI Manager or GitHub.

Why It Matters

Democratizes high-quality AI video upscaling, reducing reliance on paid tools like Topaz AI.