Home

Awesome

Noise Tools

Nodes for playing with noise. All experimental, WIP, etc.

Mix Noise is designed for generating small variations in order to create sets of similar images.

Shape Noise is designed to push noise towards (or away from) the center of the image.

Seperable Batch Noise generates noise for batches in a way that allows the seed for each item in the batch to be calculated easily.

Batch Noise Simulate generates noise for a single image that simulates the noise from a batch.

All the images below were generated using LEOSAM's HelloWorld XL 7, and the prompt was stolen from the example image.

Drag and drop any of these images into Comfy to get the workflow.

Mix Noise

This node takes two noise inputs and produces a weighted mix, optionally with a weight mask. The effect is to allow you to interpolate the noise between two extremes, or generate small variations:

Seed 10.10.30.5Seed 2
11111

Inputs

Outputs

Usage

Generating small variations

Masked noise

This is a bit more experimental!

If you leave noise2 unconnected, the effect of the mask is to reduce the noise (mixing it with zero) in the areas masked. This is similar in concept to the Shape Noise node below.

Shape Noise

This node takes a noise input and applies some simple shaping options to it. The effect is to push detail into the middle, or out of it.

0.10.20.30.40.5
+11111
-11111

The negative push breaks down when the detail is pushed out to the corners, and the sampler can't make any sense of it. The more realistic an image, generally the sooner it breaks down.

A small negative value can help detail spread across the whole image (as there is a tendency for many models to focus on the main feature in the center)

Inputs

Outputs

Concept

The sampling process works by trying to remove noise to reveal an image. It is trained with noise evenly distributed over the latent, so that is what it tries to remove. If There is more noise in one part of the image than another, it tries to make sense of that by making that part of the image 'noisier' - that is, more busy, or detailed.

So to a slight extent you can make part of the image more detailed (or more bland) by shifting the magnitude of noise around.

Usage

Seperable Batch Noise

The idea here is to generate noise for a batch of images in such a way as to give each entry in the batch a well defined seed that can be reused.

Inputs

So the third image generated had a seed of seed + 2*seed_delta

Outputs

Usage

Create noise for a batch like this. Seed 1000, delta 1.Because delta was one, the images used seeds 1000, 1001, 1002, 1003, so this will generate the second image from the batch
imgimg

Batch Noise Simulate

If you generated a batch of images (not using Seperable Batch Noise) and now want to reproduce the second image...

Inputs

Outputs

Usage

If you generated a batch of images like thisYou can get (say) the second like this (the latent is making a batch of 1, the original noise has the same seed, and batch_entry picks)
imgimg