Get Noise from a Conditioning or ConDelta

This node's simpler sibling, ConditioningGetRandom, gives you one Gaussian-noise recipe seeded and scaled off a conditioning you feed it. ConditioningGetNoise does the same basic job - generate a noise conditioning shaped and scaled against an input, deterministic for a given seed - but exposes four selectable variants instead of one, via the type field. Worth being upfront about: the pack doesn't document what actually differs between type1 through type4 beyond their labels, so this is a menu to experiment with rather than a spec to reason about in advance.

How it works

You give it a conditioning to use as a template, a strength, a type, and a seed. It produces noise scaled by strength and seeded for reproducibility, with type selecting which of the four noise generation approaches gets used. Same category of tool as ConditioningGetRandom - a control or a source of controlled jitter, not a meaningful concept in itself - just with more knobs to turn if the default recipe isn't giving you the texture you want.

The inputs and outputs that matter

• conditioning - the template your noise is shaped against.
• strength - default 1, step 0.01, controlling the noise magnitude.
• type - default type1, with type2, type3, and type4 as the other options. Since the pack doesn't spell out the difference, treat this as trial and error: generate on a fixed seed with each option and compare.
• seed - default 0, standard ComfyUI seed range, for reproducibility.

Output is a single CONDITIONING.

How to install it

Via ComfyUI Manager: search "ComfyUI-ConDelta", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/envy-ai/ComfyUI-ConDelta

then restart. No model downloads needed.

Common issues & troubleshooting

Don't expect an obvious visual difference between the four types on the first try. Since none of them are documented beyond a bare label, the practical approach is a fixed-seed comparison: same conditioning, same strength, same seed, cycle through type1 to type4, and look at what actually changes in your output. If two types look identical to you, that's useful information too.

If you just want one dependable noise source without deciding between four unlabeled options, use ConditioningGetRandom instead. It's the simpler node in the pack for the same underlying job - a seeded, scaled Gaussian noise conditioning - without the extra decision this node adds.

Same caveats as any noise node in this pack apply: it's most useful as a control for testing whether a real ConDelta is doing more than an equivalent-magnitude random push would, or as a cheap source of controlled variation layered on top of a conditioning you already like.