Get random Gaussian noise from a Conditioning or ConDelta

Not every experiment in this pack needs a "real" delta. Sometimes what you actually want is a control: if you're trying to prove a ConDelta you built is doing something meaningful, the honest test is comparing it against noise of the same rough scale, not just eyeballing whether the output changed. ConditioningGetRandom gives you that noise - seeded, so it's reproducible, and scaled off the conditioning you feed it rather than pulled out of thin air.

How it works

You pass in a conditioning (which the node uses as a template - its display name is explicit that this works on "a Conditioning or ConDelta"), a strength, and a seed. It generates Gaussian noise shaped to match, scaled by strength, deterministic for a given seed. The point isn't to modify your input conditioning's content - it's to produce a fresh noise conditioning at a comparable scale, which you can then treat exactly like a ConDelta: add it, scale it, clamp it, apply it.

The inputs and outputs that matter

• conditioning - the template your noise is shaped against.
• strength - default 1, step 0.01, no explicit min/max in the schema (unlike most of this pack's strength fields, which cap at ±100). This is your noise magnitude.
• seed - default 0, up to the standard ComfyUI max-seed range. Same seed, same strength, same shape of input, same noise every time.

Output is a single CONDITIONING - noise, not a meaningful concept, ready to plug into ConditioningAddConDelta or ConditioningScale alongside your real deltas.

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. Nothing extra to download for this node.

Common issues & troubleshooting

Use it as a sanity check, not a creative tool. The most practical reason to reach for this node is to answer "is my ConDelta actually doing something, or would equivalent-magnitude noise produce a similarly different image?" Generate with your real delta, generate with this node's noise at a matching strength on the same seed, and compare. If the noise version looks about as different from your baseline as the real delta did, your delta isn't carrying much signal.

Vary the seed for variety, keep it fixed for comparison. If you're using this to add controlled jitter to a workflow, cycle the seed. If you're using it as a control against a real delta, hold the seed fixed and only change strength, so you're comparing like for like.

It's a sibling of ConditioningGetNoise, not the same node. This one has a single Gaussian recipe; the other exposes four selectable noise "types." If one doesn't give you the texture of randomness you're after, the other is worth trying.