Dazzle Options

Dazzle Options is the advanced-configuration node for the Smart Resolution Calculator (SmartResCalc). It's not a calculator and it doesn't generate anything. It's a small control panel that tunes how SmartResCalc's spectral blending behaves - the feature that lets a noise pattern's spatial structure steer where objects land in your generated image. If you've never heard of spectral blending, you can skip this node entirely and never miss it. If you've ever seen a SmartResCalc image go abstract at high resolution, or wanted your old v0.10.4-era outputs back, this is the lever.

How it works. Under the hood, SmartResCalc's spectral blending works in the frequency domain: it takes your noise pattern (DazNoise Plasma, Gaussian, whatever you picked as fill_type), FFTs it, and replaces the low-frequency components of standard Gaussian latent noise with normalized versions of the pattern's. The result keeps the N(0,1) statistics diffusion models expect while smuggling in the pattern's structure. Dazzle Options just changes how that normalization happens. It builds a plain dict and hands it to the calculator's dazzle_options input. Connect nothing, and SmartResCalc uses its defaults - so this node is strictly opt-in, and it chains: feed one Dazzle Options into another's options_in and the downstream one wins on any field it sets.

The inputs that matter. Only a couple will ever mean anything to you:

• norm_mode - auto (default) lets SmartResCalc pick: per-bin normalization for real images, global-RMS for noise patterns. per_bin normalizes each frequency bin individually, transferring spatial structure via phase only - resolution-independent, the recommended choice for img2noise at high resolution. global_rms scales every bin by the same factor; that's the pre-0.10.4 behavior, and it can overwhelm at big sizes. Set it here when you want to reproduce older outputs exactly.
• whitening - a 0.0-to-1.0 blend between global-RMS (0.0, full amplitude transfer) and per-bin (1.0, phase-only). Only consulted in auto mode; 0.5 is half and half.
• cutoff_curve - the shape of the frequency rolloff mask. gaussian is the smooth default (avoids Gibbs ringing); sharp is a brick-wall cutoff that can ring.
• phase_randomize - randomizes pattern phases before blending; decorrelates the pattern from your sample in img2img+img2noise mode while keeping spatial structure.

Install. It ships inside the pack, so installing the pack installs this node. In ComfyUI Manager, search "Smart Resolution Calculator" (or DazzleNodes / djdarcy), click Install, restart. Command line works too:

cd ComfyUI/custom_nodes
git clone https://github.com/DazzleNodes/ComfyUI-Smart-Resolution-Calc.git

Then restart ComfyUI. There are no pip dependencies and no model downloads - requirements.txt is effectively empty, and it runs on ComfyUI's own torch, PIL, and numpy. It's MIT-licensed, by Dustin Darcy.

Gotchas. The most common "problem" here is version mismatch, not a bug: newer SmartResCalc defaults to per-bin normalization, so if your saved workflow's output subtly shifted after an update, that's why - flip norm_mode to global_rms. And remember Dazzle Options tunes normalization only; the real dial is blend_strength on the calculator itself, where Plasma patterns stay coherent only below roughly 0.15-0.17, while per-pixel noise types run happily at 1.0. If your latent looks broken, check that first. The node is otherwise remarkably boring - which, for a config node, is exactly what you want.