Legacy2_ClownOptions_SDE_Noise

Most of the time, the noise a sampler injects during SDE sampling is generated fresh from a random seed. This node lets you swap that for a specific latent of your own choosing. Instead of "random noise," the sampler pours your latent in as the noise for the early steps. It's a niche, deliberately experimental control - this is the Legacy version of the node - but a powerful one if you know what you want out of it.

The use case is steering the character of the noise rather than leaving it to chance. Because the "noise" you feed in carries structure, it can bias the generation toward a particular texture, composition, or feel - closer to a guide than to true randomness. It's the sort of trajectory-shaping tool the pack exists to expose.

How it works

You hand it a LATENT and tell it how many steps to use that latent as the SDE noise source. For those steps, the sampler's noise injection draws from your supplied latent instead of a random draw; after that step count, it reverts to normal behavior. The whole thing is bundled into an OPTIONS object the sampler reads.

Because this only affects SDE noise, your sampler has to actually be SDE sampling - eta above 0 - for the substituted noise to do anything.

The inputs and output

• sde_noise (LATENT, optional) - the latent you want used as noise. This is the payload; a VAE-encoded image, another generation's latent, or any latent-shaped source. Without it, there's nothing to substitute.
• sde_noise_steps (default 1) - how many steps to apply your latent as the SDE noise. 1 affects only the first step; raise it to extend the influence deeper into the run.

There's the usual optional options chain input, and a single options (OPTIONS) output for the sampler.

Installing RES4LYF

ComfyUI Manager: search RES4LYF, install, restart. Or clone:

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

Portable ComfyUI: embedded pip under python_embedded/Scripts/. Restart, hard-refresh with F5. No model downloads.

Common issues

Shape mismatch errors. The sde_noise latent has to match the dimensions of what you're generating. A latent encoded at a different resolution (or a different batch size) will throw. Encode/generate it at the same size as your target.

No visible effect. Either your sampler's eta is 0 (no SDE noise to replace), or sde_noise_steps is so low the substitution is over before it matters. Raise the step count and confirm eta is above 0.

It's the legacy node. The display name reads Legacy2_ClownOptions_SDE_Noise. It still works, but if you're building fresh and want the current SDE tooling, look at the non-legacy options - this one's kept for compatibility with older workflows.

Nothing reaches the sampler. Standard ClownOptions rule: the options output must land on a RES4LYF sampler's options input to have any effect.