BMAB Noise Generator

Exactly what it says: this node outputs a random noise image at whatever size you specify. Not a latent, not a sampled starting point for a diffusion run - a plain IMAGE full of noise. That distinction matters for what it's actually useful for: seeding a texture or grain layer for compositing, filling a canvas placeholder before you've decided what should go there, or feeding a downstream node that expects an image input but where the content doesn't matter yet, only the dimensions.

How it works

The default path is the two width/height widgets - pick a size between 256 and 2048 (in steps of 8, the usual diffusion-friendly alignment) and it generates noise at exactly that resolution. The two optional inputs are the more interesting part: wire in a bind from BMAB's pipeline, or a latent, and the node can presumably size itself off whatever's already flowing through your graph instead of you re-typing dimensions by hand - useful if you're generating a noise layer that needs to match a canvas BMAB Resize By Ratio or Reframe already established upstream, rather than guessing at matching numbers on two separate nodes.

The inputs and outputs that matter

• width / height (required, default 512, range 256–2048, step 8) - the manual size controls. These are what drive the output if you're not wiring in bind or latent.
• bind (optional, BMAB bind) - pull sizing context from an existing BMAB pipeline instead of setting width/height by hand.
• latent (optional, LATENT) - match an existing latent's dimensions.
• Output: image - a noise image at the resolved size.

How to install it

ComfyUI Manager: search comfyui_bmab. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/portu-sim/comfyui_bmab.git
cd comfyui_bmab
pip install -r requirements.txt

No model or extra dependency needed - this is a pure generation-from-nothing utility node.

Common issues & troubleshooting

You wired bind or latent and the width/height widgets seem to be ignored. That's the expected behavior if the optional inputs take priority when present - treat the widgets as the fallback for when nothing's wired in, not as a value you can override alongside a connected bind/latent.

You wanted a starting point for actual generation, not just an image. This node produces a plain IMAGE, not a LATENT, and it's not a substitute for Empty Latent Image or an img2img noise seed in your sampler. If you need noise to actually feed a sampling pass, you're looking for the wrong node - this one's for visual/compositing use, not for driving diffusion.

The output looks like uniform static rather than anything useful on its own. That's correct and expected - it's raw noise, not a texture or pattern generator. If you need structured noise (Perlin, fractal, etc.) for a specific visual effect, this node doesn't offer that; it's the simplest possible random fill.