Nodes/IAMCCS-nodes/IDEO Noise Debug
ComfyUI Node

IDEO Noise Debug

Verify the seed actually reached the sampler

By IAMCCS·Created 11 months ago·Updated 8 days ago· 113
IDEO Noise Debug
  • noise
  • noise
  • report
labelideogram_noise

"Same seed, different result" is the single most confusing bug in diffusion workflows, and IAMCCS IdeoNoiseDebug exists to tell you which half of that sentence is a lie. It's a passthrough probe that sits between your RandomNoise node and the sampler, reads the seed the noise object is actually carrying, and prints it. No mutation, no side effects - it only reports.

The author's docstring nails the use case: "Put it between RandomNoise and the sampler. It does not change the noise object; it only reports the seed held by RandomNoise so storyboard clone reports can separate prompt issues from seed/noise issues." In a multi-queue storyboard pipeline, that separation is everything - you need to know whether the divergence came from a prompt edit or from the noise silently changing.

What it does

Required input: noise (the NOISE type - anything RandomNoise or a custom noise node emits). Optional label (default ideogram_noise) tags the probe so you can distinguish several in one graph. It's an output node, so the report shows up in the UI text area.

Outputs: noise passes through byte-for-byte, and report is the JSON string with the seed and whatever else the noise payload holds. Wire noise straight into your KSampler / SamplerCustomAdvanced and read report from the UI.

When you actually need it

The "clone report" context is the giveaway that this is aimed at IAMCCS's own storyboard pipeline, where the same frame gets re-generated across queues and any seed drift produces a subtle frame-to-frame mismatch you'll chase for hours. A probe makes it a ten-second check: same seed in the report = the noise isn't your problem, look at the prompt/conditioning. Different seed = you've got a caching or node-order issue, not a creative one.

It's also just a nice general sanity check if you're paranoid about whether a workflow genuinely honors a fixed seed - throw one in, run twice, compare.

Install

It ships with IAMCCS-nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

or via ComfyUI Manager → search IAMCCS, then restart. Nothing extra to install.

The caveat

The seed a noise node holds and the seed you typed can differ when a workflow randomizes or when a node forks the noise - the report tells you the ground truth, which occasionally means revealing that your "fixed" seed wasn't fixed. That's a feature, not a bug. Leave the probe inline if you want; it costs a line of UI text per run.

CategoryIAMCCS/Cine/Ideogram/Debug

Inputs (2)

NameTypeDefaultDescription
noiseNOISE
labeloptSTRINGideogram_noiseOptional label printed in the debug report.

Outputs (2)

NameTypeDescription
noiseNOISE
reportSTRING