Gigapixel AI

This node is not a new upscaling model. It's a remote control for Topaz Gigapixel AI, the commercial desktop upscaler you already paid for. You hand it an IMAGE batch, it writes the files to disk, calls gigapixel.exe from the command line, waits for the upscale to finish, and hands the results back to ComfyUI as tensors - plus the file paths, so you can see exactly where Topaz wrote them. No API key, no cloud round-trip, no Stable Diffusion weights involved. The "AI" lives inside your Topaz install; this node is mostly plumbing, and the author's own README credits choey's Comfy-Topaz wrapper as the template it's built on.

Why reach for it when Real-ESRGAN and SeedVR2 are free? Honest answer: Topaz's edge is the boring stuff. It's a proven quantity for batch upscaling and print work - people push stacks of images to print resolutions (300 DPI, 10k pixels wide) in minutes, and its noise and compression cleanup is consistent and clean. If you're already paying for the subscription, this kills an export/import hop: generate, upscale in-graph, and keep the whole pipeline reproducible in one workflow. If you're not on the subscription, this node does nothing for you - the open-source rungs (ESRGAN for pure pixels, SeedVR2 for detail) are the cheaper and often better call. This is a convenience wrapper for people with a license, not a reason to buy one.

Before you install anything: you need a licensed Topaz Gigapixel AI 8 install, and the pack needs GigapixelAI 7.3.0 or newer. gigapixel.exe is a Windows binary, so this whole pack is a Windows-only ride. No Linux, no Mac, no apologies.

The inputs that matter:

• images - your IMAGE batch. Anything from a VAE decode, an img2img pass, a loaded file.
• scale - 1 to 16, default 2. The upscale factor. 2 is usually plenty; past 4 you're asking Topaz to invent more than it resolves.
• gigapixel_exe - defaults to C:\Program Files\Topaz Labs LLC\Topaz Gigapixel AI\gigapixel.exe. This is the pack's one documented gotcha: if your install isn't in the default location, fill this in manually or the node can't find the executable.
• model - wire in a GigapixelModelSettings node to pick which of Topaz's nine models does the work.
• upscale - wire in a GigapixelUpscaleSettings node to control sharpen, denoise, compression, face recovery, and pre-downscaling.
• use_system_command and no_temp - plumbing toggles for whether it shells out via system command and whether it keeps temp files. Leave them at their defaults until something misbehaves.

Outputs: IMAGE (the upscaled batch - wire it to a Save Image), image_paths (the files Topaz wrote), and settings (a string list describing the run, handy for logging what actually happened).

Install is the standard custom-node dance, and the README lists no pip dependencies - the heavy dependency is the Topaz install itself:

cd ComfyUI/custom_nodes
git clone https://github.com/sh570655308/ComfyUI-GigapixelAI

Then restart ComfyUI. Easier: ComfyUI Manager → search "ComfyUI-GigapixelAI" and install from there.

Where people get burned: the wrong exe path (fix: set gigapixel_exe), and Topaz's CLI misbehaving silently until you've launched the app once so it can activate or log you in. Do a first run in the GUI before blaming the node. And it's slow - this is a full commercial upscaler doing real per-image work, so budget seconds-to-minutes per image. The pre-downscaling setting is the lever if you need it faster.