DisableNoise_motorway_edition

Diffusion normally starts from a random noise tensor. DisableNoise is the node that says "actually, don't." Feed it where the sampler expects noise and it hands back an empty noise tensor, so the denoiser starts from the latent you give it rather than a fresh pile of static. That's the whole trick behind img2img-style refinement, inpainting, and any workflow where the initial noise injection would just fight what you already have.

This _motorway_edition version is the same core node wrapped to live on the Motorway pipe system from agilly1989/ComfyUI_agilly1989_motorway. If you've never met the Motorway: it's a named-key "bus" that carries values between nodes under string keys. You start it with MotorwayStart, stash values on it with Motorway AxB ramps, and the motorway-ed clones are ordinary ComfyUI nodes that read their inputs from a key and write their output to a key instead of passing tensors directly.

DisableNoise is the simplest clone in the batch because the original takes no inputs at all. All it does is produce an empty noise object and store it on the bus.

What the inputs and output mean

The schema is refreshingly short:

• MOTORWAY 🚌💨 (optional) - the pipe. If you leave it unplugged the node still works; there's just no bus to write to.
• OUTPUT_NOISE_key - which key the empty noise object gets stored under. Default is NOISE. Whatever key you pick here is what you reference downstream when you want to pull that noise back off the bus with a Motorway 0x1 ramp (or just let another motorway-ed node read it).

The only output is the MOTORWAY pipe itself. In Motorway world, a node like this never hands you the noise directly - it writes it to the key you named, and you retrieve it later.

Where it fits

Custom sampling. In a plain ComfyUI workflow you'd wire DisableNoise into the noise input of KSamplerCustom/KSamplerCustomAdvanced when you want zero initial noise - classic for inpainting passes or for "clean" img2img where the input latent already carries the image and you only want the scheduled sigmas to do the work. On the Motorway, the flow looks like: encode your image → store it under a key → run a motorway-ed sampler whose INPUT_noise_key points at the key DisableNoise wrote.

One subtlety worth knowing: this node doesn't remove noise from the schedule - the sigmas still run. It only removes the random tensor that would be injected as the starting point. Keep that straight and you won't be surprised when an image still visibly denoises.

Install and the honest caveat

Install is the same for every node in this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Then restart ComfyUI. You can also search "ComfyUI_agilly1989_motorway" in ComfyUI Manager. No extra Python packages, no model downloads.

Before you spend an hour hunting for it in the node menu, one thing. The pack's own pyproject.toml describes the latest release (1.1.7) as a "HOTFIX - REMOVED MONKEYPATCH AND GENERATED NODES," and in __init__.py the generated clone nodes are registered but commented out. The author broke the clone generator against newer ComfyUI core and hotfix-disabled it - the core Motorway nodes (MotorwayStart, ramps, primitives) still work fine, but the _motorway_edition clones like this one may not appear in your menu until that gets re-enabled. It's a beta pack with a clear "if things break it's because I broke it" sign on the door. If you're learning the Motorway system, this node's job is trivial anyway; the interesting part is the key naming, and you can practice that with a plain ramp.