Advanced Consensus Configuration

Consensus conditioning is the pack's technique for making multi-image and multi-resolution conditioning more robust: instead of trusting one encode of one image at one resolution, you encode several variants and blend them so outlying noise gets rejected. UC_AdvancedConsensusConfiguration is the config node that controls that machinery - and specifically, the "Advanced" bit is resolution_samples and sample_offset, two knobs the base Text Consensus Blend Configurator doesn't have. This node's output is a config object, not conditioning; it feeds the pack's consensus-capable encoders like UC_AdvancedVisConEncoder.

What it does

It's a big schema, but the defaults are sane, and the blend_preset combo is the governor: baseline (default) for the pack's recommended config, off to bypass consensus entirely, or custom to expose all the manual parameters. For 90% of use, you'll pick a preset and touch at most one or two things. The manual knobs, when you need them:

• blend_method - consensus aligns prompts and filters noise (the fancy one), linear just averages.
• consensus_type - median rejects up to 50% of outlying noise; mean smooth-averages.
• alignment_method / alignment_threshold / similarity_threshold - how prompts are aligned before blending: by semantic similarity or by index.
• power_alpha (default 2) - soft-masking exponent; higher values penalize outliers.
• diversity_beta (default 0) - raise toward 1.5 to dampen hyper-frequent details and boost variety.
• rescale_norm (default on) - keeps activation energy high so blending doesn't wash out colors.
• global_scale, dynamic_similarity_contrast, soft_comfort_bandpass, position_weight, preserve_common_prefix - the fine print. Leave most alone at first.

Then the two features that justify "Advanced":

• resolution_samples - the exact number of adjacent resolutions sampled (odd, 1–15, default 1). A value of 1 stays one sample; 3 means the encoder renders the conditioning at three neighboring resolutions and blends them. This is the "consensus across resolution" part.
• sample_offset - the equivalent-square distance between adjacent samples in pixels (32–512, default 32). Bigger offsets produce more distinct visual grids at greater scale separation.

The output is ADVANCED_CONSENSUS_CONFIG, which you pass into a consuming encoder. The tooltip's wording - "Exact odd adjacent-resolution sample count used only when consensus is enabled" - is the gotcha to remember: resolution_samples only matters when blend_preset isn't off. Turn consensus off and these knobs do nothing, which confuses people more than anything else in this node.

Install

ComfyUI Manager → search ComfyUI-UtilsCollection → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-UtilsCollection

Restart. Deps: opencv-python and typing-extensions.

When you'd reach for it

When you're building an Advanced Visual Consensus encoder workflow and want control over how many resolution variants get blended and how far apart they sit. It's a power-user node - the "Advanced" in the name is earned - and the sensible path is: start with baseline, then experiment with resolution_samples = 3 and sample_offset = 32 before touching the consensus math underneath.