LatentNoiseBatch_perlin

Gaussian noise is static - every pixel independent, no relationship to its neighbors. Perlin noise isn't: it's built from overlapping gradients, so it comes out in soft blobs and streaks instead of a uniform hiss. LatentNoiseBatch_perlin builds you a fresh latent filled with that kind of noise instead of the flat random field ComfyUI's Empty Latent Image gives you by default.

That distinction matters more than it sounds like it should. Community experience with alternative noise types (perlin, fractal, pink, brown, and plain Gaussian) is that each biases the eventual image differently - sometimes composition, sometimes fine detail, rarely nothing. This node is a drop-in way to try the structured end of that spectrum instead of accepting whatever your sampler's default noise gives you.

How it works

Unlike the pack's other noise nodes, this one doesn't take an existing latent input - it builds one from scratch, the same way Empty Latent Image does, from width, height, and batch_size. The difference is what fills it: Perlin noise instead of blank/random. detail_level controls how fine-grained the pattern is - lower values give broader, smoother blobs; higher values pack in more texture. Because it's a straight swap for Empty Latent Image, you can drop it into any workflow at the exact point you'd normally place that node.

The inputs and outputs that matter

• width / height (default 1024×1024) and batch_size - the same sizing controls you'd set on Empty Latent Image.
• detail_level (range -1 to 1, default 0) - the one knob worth actually turning. Push it toward 1 for a busier, higher-frequency noise pattern; toward -1 for broader, smoother blob structure.
• seed - reproducibility, as usual.
• Optional details (SIGMAS) - schedule detail_level across steps instead of a fixed value, if you want the noise texture itself to shift over the course of generation.
• Output: latent - feed this directly into your sampler in place of Empty Latent Image's output.

How to install it

Via ComfyUI Manager, search "RES4LYF" and install. Or manually from ComfyUI/custom_nodes:

git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt

Activate a venv first if you're using one; on portable installs, swap pip for python_embedded\Scripts\pip.exe. Restart ComfyUI once it's in place.

Common issues & troubleshooting

The output barely differs from plain noise. detail_level at its default (0) is a mild effect - this isn't meant to be a dramatic style filter, it's a subtle bias on where the sampler's early steps have somewhere to "land." Push detail_level toward the extremes to actually see the blob structure show up in the final image, and compare against a plain Empty Latent Image at the same seed to isolate the effect.

You wanted this to feed the sampler's SDE noise instead. It doesn't touch that - RES4LYF's own samplers have separate noise_type_init/noise_type_sde settings (which include their own pyramid- and fractal-style noise types) for noise injected during sampling. This node only replaces the starting latent, before the first step, which is a different point in the pipeline.

Resolution doesn't match the rest of your workflow. Since this node builds its own latent rather than reading one, double check width/height/batch_size line up with whatever conditioning or ControlNet inputs you're pairing it with - a mismatch here is a plain shape error, not anything specific to Perlin noise.