latent_Image2Noise

Pure random noise has no structure - every pixel is independent of its neighbors. latent_Image2Noise generates noise that isn't quite that: it's derived from a source image, so it can carry a faint echo of that image's rough shapes and colors instead of being uniformly random. That makes it a useful building block anywhere you want "structured" noise rather than pure static - a grain effect that still reads as related to the picture underneath it, or a textured starting point for further processing.

How it works

image is the source, and noise_strenght (0–1, default 1 - note the field is literally spelled that way in the node) controls how strong the noise effect is overall. noise_size (0–1, default 0.1) shapes the grain size - smaller values presumably produce finer, tighter noise; larger ones coarser blobs. color_noise (0–1, default 0) controls how much of the source image's color bleeds into the noise versus keeping it grayscale, and saturation, contrast, and blur shape the resulting texture further. There's also an optional mask input, with mask_strength, mask_scale_diff, and mask_contrast controlling how that mask is applied - so you can restrict or blend the noise effect to specific regions of the image rather than the whole frame.

Inputs and outputs

• image (required) - the source picture.
• noise_strenght - overall noise intensity.
• noise_size - grain size.
• color_noise - how much source color leaks into the noise vs. staying grayscale.
• saturation, contrast, blur - texture shaping.
• mask (optional) - restrict where the noise applies.
• mask_strength, mask_scale_diff, mask_contrast - how the mask itself is applied.

One output: IMAGE - the resulting noise-ified picture.

How to install it

ComfyUI Manager → search ComfyUI-Apt_Preset, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

then run install.bat for dependencies (Windows-only - on Linux/macOS run its pip install lines by hand) and restart ComfyUI.

Common issues & troubleshooting

Don't confuse this with latent_chx_noise, this pack's other node in the same category - that one is a general noise/grain effect applied to an image with channel and color controls; this one is purpose-built to derive noise specifically from a source image's content, with knobs (color_noise, mask_scale_diff) that only make sense in that context. If you just want quick film grain, the other node is simpler; if you want the noise to actually track the source image's shapes and color, this is the one.

Output looks like pure static with no relation to the source. Check color_noise isn't left at its default of 0 (grayscale-only noise, no color bleed from the source) and that noise_size isn't set so small that any structure gets washed out - both push the result toward looking more like plain random noise.

Mask isn't limiting the effect the way you expected. Confirm mask_strength isn't at a value that makes the mask barely register, and that the mask itself actually covers the region you intend - mask_scale_diff and mask_contrast reshape how the mask edges are applied, so an unexpected result there is usually a mask-tuning issue rather than the core noise generation misbehaving.

IMPORT FAILED on startup. This pack has at least one reported import failure in the wild on a stock install, generally traced to skipping install.bat's dependency step - which has to be run manually off Windows. Check the console for the actual missing package before reinstalling.