ImageNoiseBytes

The name is doing real work here. ImageNoiseBytes from the Allor Plugin doesn't sample a fancy distribution - it pulls actual random bytes from the OS and turns them into noise. No bell curves, no shape parameters, no bias knobs. Just raw, uniform, maximum-entropy static. In a noise family that's all about personality (Beta skews, Binomial dots, Gaussian grain), Bytes is the honest one: it's the closest thing to pure television static you can add to an image, and sometimes that's exactly the effect you want.

How it works

The mechanism is almost embarrassingly simple: np.random.bytes() fills a buffer with uniformly random byte values, which are reshaped to the image's dimensions, divided by 255, and added to the selected channels (subtracted if invert is on). Because the bytes are uniform over the full 0–255 range, every brightness level is equally likely - the noise has no texture to it beyond the pixel level. It's the "snow" of an untuned channel, applied per frame.

Like every node in the Allor noise family, it routes through the shared channels selector (rgb, rgba, rg, rb, ra, gb, ga, ba, r, g, b, a), so you can restrict the static to specific channels - a alone textures the alpha, g alone adds green-channels-only interference, which is a classic old-green-screen look. And monochromatic (true/false) flips between one shared noise pattern across all channels (monochrome static, like an old TV) and independent noise per channel (color confetti). With no strength or shape inputs at all, those two switches plus invert are your entire toolbox.

The honest trade-off: because the bytes are pure uniform random, this is the harshest of the four noise nodes by default. It's all-or-nothing full-range static. If you want a subtle film grain you want Gaussian; if you want controllable character you want Beta or Binomial. Reach for Bytes when you genuinely want loud, dirty, full-spectrum noise - glitch effects, "corrupted data" horror transitions, masking prep where you need maximum entropy in a region.

Inputs and outputs

• images (IMAGE) - single image or a whole batch; each frame gets independent noise.
• monochromatic (false/true) - shared vs per-channel noise.
• invert (false/true) - add vs subtract.
• channels - which channels receive the static.

Output is an IMAGE at the same resolution and channel count, alpha preserved.

Installing it

It ships in the Allor Plugin: ComfyUI Manager → search "Allor Plugin" → install → restart. Manual clone if you prefer:

cd ComfyUI/custom_nodes
git clone https://github.com/Nourepide/ComfyUI-Allor

No model files needed. First launch writes config.json (daily auto-update on by default) - editable if you want to change update behavior.

Where people get burned

Two things catch people. First, there's no volume control, so if the full-range static is too loud, this isn't the node to fight with - drop to ImageNoiseGaussian and dial strength instead. Second, since it's true random bytes, the noise is completely independent per frame: in an animation you'll get a full static flicker, which is great for a "broken signal" scene and wrong for anything stable. And the classic family trap applies - leaving invert on quietly turns your snow into a subtraction pass.