LatentNoised

LatentNoised takes a latent and adds noise to it, with far more control than "denoise 0.6" gives you. That's the short version. It's the node you use when you want to choose what noise goes into an img2img or unsampling pass - the type, the strength, the seed, whether it's masked to one region - instead of letting the sampler quietly handle it.

Why bother? Because in RES4LYF's world, noise is a tool you shape on purpose. The pack is built around the idea that modulating noise and sigmas over time buys real quality, and a lot of its more interesting workflows (unsampling, resampling, latent-guided img2img) depend on injecting a specific kind of noise at a specific place rather than the generic Gaussian a KSampler adds for you. There are 17 noise types here - plain gaussian, fractal, uniform, and a family of pyramid variants - and different types give the sampler different textures to work from. For most people this is a specialist node, but if you're following a RES4LYF img2img recipe that says "noise the latent first," this is what does it.

How it works

You hand it a latent, pick a noise type and strength, give it a seed, and it returns the same latent with that noise mixed in. Optionally you can supply your own pre-made noise as a latent, or a mask so the noise only lands where you want it.

The inputs and outputs that matter

• latent_image (LATENT) in, latent_noised (LATENT) out - the noised latent, ready for the next stage.
• noise_type - the big one, 17 options. gaussian is the safe default; fractal and the pyramid-* variants give structured noise that behaves differently under sampling. Worth experimenting when a plain result looks flat.
• noise_strength (default 1) - how much noise to add. This is your main dial; lower keeps more of the original latent, higher buries it.
• noise_seed - fixes the noise pattern so a result is reproducible.
• normalize (default false) - renormalize after adding noise; flip it on if strengths above 1 start blowing out.
• Optional mask (MASK) - confine the noise to a region. Optional latent_noise (LATENT) - supply your own noise instead of generating it. alpha and k shape the fractal/pyramid noise families and are safe to ignore until you're deliberately tuning them.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Manual:

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

then restart and hard-refresh (F5). Nothing to download.

Common issues & troubleshooting

You noised a latent and then let the sampler add noise too. That's double noise and it looks like it. If you're pre-noising for an unsampling/resampling flow, set your RES4LYF sampler's mode accordingly so it isn't also injecting noise on top - that's the exact reason the pack's samplers have unsample/resample modes that disable internal noise addition.

The mask did nothing. Masks here are soft - a flat white mask noises everything, a partial mask feathers it. If you expected a hard boundary, check the mask you fed in actually has the shape you think.

Strengths above 1 wreck the image. That's the sign to turn on normalize, which rescales after the add so high strengths don't clip.