SonarBlendedNoise

Most sonar noise nodes are one-trick ponies: this flavor, that distribution. SonarBlendedNoise is the glue that turns them into a palette. It takes two custom noise items, mixes them at a ratio you control, and hands back a single SONAR_CUSTOM_NOISE chain. Pure gaussian is boring and pure voronoi is chaos; a 75/25 mix of gaussian and pyramid is often exactly what a generation wants. This is the node you reach for when no single noise type is doing the job.

The controls are blessedly few:

• custom_noise_1 / custom_noise_2 - the two noise items to blend, each accepting a SONAR_CUSTOM_NOISE chain.
• noise_2_percent - the blend ratio, 0.5 default. This is the knob you'll actually be turning. Two useful behaviors to know: at exactly 0 the second input is optional and isn't even called, and at exactly 1 the first isn't - so going from 0.000000001 to 0 is a real, structural difference, not a rounding artifact.
• blend_mode - lerp (default), plus inject and subtract_b. If ComfyUI-bleh is installed you get its whole blend-mode library here too, which is where this gets interesting.
• custom_noise_mask - the advanced path. Connect a third noise item and it becomes a spatial mask: its output is normalized to 0–1 and drives the blend ratio per-pixel instead of the global noise_2_percent. The mask is then clamped to 0–1, and noise_2_percent adds to it first - so you can force the effective range to 0.5–1.0 by setting it to 0.5, or 0–0.5 by setting it to -0.5. Note that once a mask is connected, both noise inputs become mandatory.
• normalize - default / forced / disabled output normalization. The tooltip suggests forced for weird blend modes that stretch the range.
• factor / rescale - the pack-wide strength pair.

Output is SONAR_CUSTOM_NOISE, chainable onward, and consumed wherever custom noise is accepted. Install is the standard one-liner:

cd ComfyUI/custom_nodes
git clone https://github.com/blepping/ComfyUI-sonar

The practical pattern: build two chains you like, blend at 50/50, then sweep noise_2_percent across a few seeds. Because custom noise only shows up on samplers that inject noise, run this through Euler a, DPM++ SDE, or a sonar momentum sampler. And the usual pack caveats apply - noise placement between GPU and CPU is inconsistent, so seeds aren't perfectly portable across hardware, and the pack's inputs shift between versions. None of that stops it from being the node that makes the whole "noise toybox" concept cohere: every other noise node is an ingredient, and this is the mixing bowl.