Colored Noise (Initial)

Colored Noise (Initial) is the small, quiet node in the akashzeno/ComfyUI-ColoredNoiseDiffusionSampling pack - and that's fine, because it does one specific thing well. It outputs a NOISE object that produces colored initial latent noise: your starting latent, instead of being plain white Gaussian, is shaped to a single constant color. Wire it into the noise input of SamplerCustomAdvanced (the NOISE output is duck-typed to that socket, exactly like ComfyUI's built-in noise nodes).

Where the pack's other two nodes color the noise injected at every sampling step, this one only colors the start. That's the whole job. No trajectory, no frequency sweep - one fixed spectrum for the latent you begin from.

The inputs

• noise_seed - the seed for the initial latent. Kept separate from any sampler seed, so it behaves like ComfyUI's own initial-noise seed.
• color - the preset: white, pink, brown, blue, violet, or custom. Those map to spectral exponents: pink = 1, brown = 2, blue = -1, violet = -2. Positive exponents push energy into low frequencies (pink/brown - softer, more low-frequency texture), negative exponents push it high (blue/violet - sharper, spikier).
• alpha - used only when color = custom; pick any exponent in [-8, 8] for a color the presets don't cover.
• energy_scale - the "heat" knob, default 1.0. It scales the noise's standard deviation after the unit-variance renormalization, so values away from 1 deliberately change how loud the noise is.

Output: a single NOISE.

Why you'd reach for it

Two reasons, really. First, it's the cheapest way to feel out the pack's coloring idea without touching your sampler: plug it in, try pink, try blue, see if the start color changes anything you care about. Second, some people genuinely prefer a colored start - a pink-noise start is a mild "softer composition" nudge that some models respond to. It also honors batch_index determinism (a subset of a batch stays bit-for-bit consistent with the full draw) and handles nested video latents, so it plays nice with batches and video workflows.

Install

No dependencies beyond what ComfyUI already has. Via ComfyUI Manager, search "Colored Noise"; or:

cd ComfyUI/custom_nodes
git clone https://github.com/akashzeno/ComfyUI-ColoredNoiseDiffusionSampling

Restart ComfyUI - it appears under sampling/colored_noise. You'll see [ColoredNoiseDiffusionSampling] startup logs, and each generation logs a line like colored initial noise: alpha=-1.00 | energy=1.00 | seed=... so you can confirm it's actually running.

The one thing to remember

If you actually want the per-step colored noise treatment - the part where the spectrum shifts from broadband early to high-frequency detail late - this node isn't it. That's the job of ColoredNoise_SamplerSelect (a SAMPLER for the same custom-sampling graph) or the all-in-one ColoredNoise_KSampler. The initial-noise node is a constant-color start, not a substitute for the real thing. Think of it as the appetizer.