LatentNoiseBatch_gaussian

The plain-vanilla noise generator in RES4LYF's noise toolkit, sitting next to the pack's fractal (colored) noise sibling. Where fractal noise has frequency structure baked in, this is standard gaussian noise - the same kind of randomness a sampler normally starts from by default - except exposed as its own node with mean and std directly under your control, instead of always drawing from a fixed, zero-centered distribution.

Most workflows never touch this and don't need to: ComfyUI's built-in noise generation is gaussian with mean 0 already. Where this node earns its place is when you want to deliberately bias the starting point - a nonzero mean nudges the whole latent toward one direction before the sampler ever runs a step, and adjusting the standard deviation changes how strong or subtle that initial randomness is, which in turn affects how much room the sampler has to diverge from run to run.

What you set

• latent (LATENT) - required, the latent shape the noise is generated for.
• mean (FLOAT, default 0) - shifts the center of the noise distribution. Zero is standard; nonzero deliberately biases the starting point in one direction.
• std (FLOAT, default 1) - the spread of the noise. Standard is 1; lower values mean gentler, more subtle initial randomness, higher values mean wilder variation between seeds.
• steps (INT, default 0) - most likely sets how many independent noise draws go into the output batch - consistent with the "Batch" in this node's name, and with the pack's other batch nodes like Latent Batcher, though the pack doesn't spell this out explicitly.
• seed (INT) - reproducibility, same as any sampler seed.
• Optional means, stds, steps_ (SIGMAS) - schedule any of these across steps rather than holding a single fixed value, the same pattern RES4LYF uses throughout its noise and latent-adjustment nodes.
• Output: LATENT - the generated noise batch, ready to feed into a sampler in place of default noise.

Installing it

Comes with the full RES4LYF pack:

• ComfyUI Manager - search "RES4LYF," install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF, then cd RES4LYF && pip install -r requirements.txt, restart.

No extra downloads for this node.

Common issues

No visible difference from default sampling. At mean=0, std=1 this is functionally the same as standard gaussian noise - you won't see a difference until you actually change one of those two values. It's a starting point for deliberate experimentation, not something that improves results just by being present.

Pushing std too far breaks the image. A standard deviation much above 1 gives the sampler more randomness to work with than it was tuned to handle, which tends to show up as instability or incoherence in the result. Move it gradually rather than jumping to an extreme value.

Unsure what mean should be set to. There's no universal right answer here - a nonzero mean is a deliberate bias, and what it does to your specific image depends on your model's latent space. Treat it as an experimental knob: nudge it a little, generate, compare against mean=0, and judge from the actual result rather than expecting a documented target value.