⚡ Gigaresolution (RTX Super Res)

Gigaresolution (RTX Super Res) upscales your images with NVIDIA's RTX Video Super Resolution - the same tensor-core hardware path that powers NVIDIA's video upscaling - inside ComfyUI. Feed it an image, get back a bigger one, usually in a fraction of a second. It's the "more pixels, on an RTX card, want it now" answer, and it's the node in this pack with the heaviest hidden prerequisites, so let's get those out of the way first.

The three requirements nobody puts on the box

The pack README breezily claims "No extra dependencies required." The RTX node is the exception that makes that claim a lie. To actually run it you need all three:

1. An NVIDIA RTX GPU (the tensor cores do the work; no RTX, no node).
2. The Python package, which the node's own error message tells you to install:

pip install nvvfx

3. The NVIDIA Video Effects SDK installed on your system, from the NVIDIA developer site.

If nvvfx is missing, the node raises a clear error with exactly those install instructions - at least you'll know. If the SDK is missing, you'll hit a different failure downstream, which is the more confusing of the two.

How it works

The node runs nvvfx.VideoSuperRes per frame at your chosen quality level, snapping the output dimensions to multiples of 8 and batching internally to stay under a 16-megapixel ceiling. Three inputs:

• images (IMAGE) - connect your VAE-decoded output here.
• scale (default 2, range 1–4) - multiplier; 2.0 doubles resolution.
• quality - LOW / MEDIUM / HIGH / ULTRA, default ULTRA. Higher quality costs more GPU time; the difference is real but modest on clean sources.

The single output is upscaled_images, which you can pass on to a save or another processing node.

The expectation-setting paragraph

This is where the KB's upscaling essay earns its keep. RTX VSR is an interpolator, not a generator - the community has spent hundreds of comments establishing that it's "more like Lanczos, it's an interpolation method, not generative." That's not a flaw; it's the reason people like it. On an already-clean image it produces natural-looking pixels fast, without the smoothed-out plastic feel that generative upscalers sometimes leave. But it will not add pores, eyelashes, or fabric weave, and it cannot invent a face. If your job is "more detail on a soft image," you want SeedVR2 or a tile-ControlNet workflow instead. If your job is "clean upscale, now," this is the one.

Install the pack

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/comfyui-gigachad.git

or search comfyui-gigachad in ComfyUI Manager, then restart - then do the pip install nvvfx + SDK step above.

Verdict: genuinely useful if you're on RTX hardware and want the fastest clean upscale available, and it's a rare in-ComfyUI wrapper for this exact hardware path. Just read the three requirements twice, and match it to the "more pixels" job - not the "more detail" one.