SONAR_CUSTOM_NOISE to NOISE
The bridge that lets stock samplers use the sonar noise toys
- custom_noise
- NOISE
The whole ComfyUI-sonar noise collection speaks one language: the pack-internal SONAR_CUSTOM_NOISE type. Stock ComfyUI's custom-sampling pipeline speaks another: NOISE. SONAR_CUSTOM_NOISE to NOISE is the translator, and it's how you use the pack's noise types without adopting a Sonar sampler at all.
Why you'd reach for it
SamplerCustomAdvanced takes a NOISE input on its noise socket - that's how custom noise gets injected into a normal custom-sampling graph. If you've built a workflow around a stock sampler and you want colored or structured starting noise, this node is the bridge: build a SONAR_CUSTOM_NOISE chain on one side, feed it in here, and plug the resulting NOISE into SamplerCustomAdvanced (or any other custom sampler that takes a NOISE input). You get the toybox without having to trust the pack's samplers.
The inputs
custom_noise- yourSONAR_CUSTOM_NOISEchain. This is the thing you're converting.seed- for the generated noise. Note this node has a realseedwidget, unlike the author's deliberate rename inSonarCustomNoiseParameters.cpu_noise- defaults totrue(noise on CPU). Flip it and noise generates on GPU, which is faster but - per the pack's README errata - can change seeds across GPU models. If you care about reproducibility across machines, leave it on CPU.normalize- defaulttrue, normalizes the noise to 1.0 strength.multiplier- a flat multiplier applied after everything else. Setting it to 0 generates no noise at all, which is the node's built-in "disable noise" switch.
The one thing that will bite you
The node description carries the warning: it doesn't work with noise types that depend on sigma. That means anything whose output varies with the current noise level - Brownian, ScheduledNoise, and friends - will not behave when converted through this node. The reason is structural: the converter has to produce a static NOISE tensor up front, but sigma-dependent types need to know the schedule step before they can generate. So if your converted noise looks wrong or static in a way you can't explain, check whether your chain includes a sigma-dependent type and swap it for a static one (Gaussian, colored noise, pyramid, etc.).
How it fits the wider picture
This is the node to reach for when you want the noise part of the pack but not the sampler part. The README notes that connecting a custom noise node to a Sonar sampler overrides the sampler's noise type - that's the all-in path. This node is the pick-and-choose path: your stock sampler, their noise.
Install
ComfyUI Manager → search "ComfyUI-sonar" → Install → restart, or:
cd ComfyUI/custom_nodes
git clone https://github.com/blepping/ComfyUI-sonar
Then restart. No models, no extra dependencies. Same standing caveat as everything in this pack: node inputs shift between versions, and noise seeds can differ between GPU models - pin your git revision if you're chasing a specific result.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| custom_noise | SONAR_CUSTOM_NOISE,OCS_NOISE | Custom noise type to convert. The following input types are supported: SONAR_CUSTOM_NOISE, OCS_NOISE | |
| seed | INT | 00–18446744073709550000 | Seed to use for generated noise. |
| cpu_noise | BOOLEAN | true | Controls whether noise is generated on CPU or GPU. |
| normalize | BOOLEAN | true | Controls whether generated noise is normalized to 1.0 strength. |
| multiplier | FLOAT | 1.000-10000–10000 | Simple multiplier applied to noise after all other scaling and normalization effects. If set to 0, no noise will be generated (same as disabling noise). |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| NOISE | NOISE | — |