🎈LG_Noise

Most noise in ComfyUI happens in latent space, invisibly, inside the sampler. LG_Noise is the opposite: it makes noise you can see and control, and it can paint that noise onto a real image. Film grain, film-dissolve transitions, texture you can aim at specific regions - it's the pack's tool for "pretend this was shot on film" and for img2img preprocessing that isn't the sampler's business.

The README calls it a "噪波节点" (noise node) for pre-processing images, and it's one of the few nodes in this pack with real depth to it.

How it works

Three noise types, all generated in image space:

• fade - a straight blend: image × (1 − opacity) + colored_noise × opacity. Even, wash-in noise over the whole frame.
• dissolve - the old video-edit effect: random pixels get replaced by noise, controlled by both opacity and density. Dense and high-opacity is a hard grain storm; low values give you sparse sparkle.
• gaussian - additive noise: image + noise × opacity × 2, clamped. The subtle one for grain, because it textures rather than replaces.

Then the modifiers:

• strength (1–32) - above 1, it generates noise at a reduced resolution and upscales it with nearest-neighbor, producing chunky structured blobs instead of fine grain. This is the "make it look like big soft noise shapes" control.
• density (0–1) - how much of the frame actually gets noise. Sparser, not subtler.
• sharpen (−32..32) - blurs or sharpens the noise component after blending. Positive sharpens the grain texture; negative smears it.
• brightness (0–3) - multiplies the noise's brightness.
• color / random_color - random_color on gives per-pixel RGB noise; off gives a single-color noise in whatever hex you set in the LG color picker.
• seed - −1 means random every run. Set a number for reproducibility.

The two optional inputs are where it gets useful: image_optional feeds the image to noise (leave it empty and it generates a standalone 512×512 noise canvas), and mask_optional confines the noise to the mask's region - grain on the subject only, clean background.

Output: IMAGE.

When to reach for it

• Film grain over a render - gaussian, low opacity, seed fixed, done.
• Transition frames or glitchy texture - dissolve.
• Stylization input for img2img - noise-conditioned images can nudge a model toward texture-heavy outputs.
• Procedural backgrounds - no image in, tune strength/color, and it's a noise-texture generator.

Installing it

Part of Comfyui_LG_Tools - Manager (search "Comfyui_LG_Tools") or:

cd ComfyUI/custom_nodes
git clone https://github.com/LAOGOU-666/Comfyui_LG_Tools.git
pip install -r requirements.txt

Restart, then right-click → 🎈LAOGOU → Utils.

Troubleshooting

• "Why is it different every run?" - seed −1 rolls a new seed each execution (it even prints the seed to the console). Pin a seed for reproducible grain.
• Grain looks like big blocks - strength above 1 creates chunky upscaled noise intentionally. Drop strength to 1 for fine grain.
• Nothing shows - check opacity, and remember density is a separate gate: with density 0, no pixels get noise regardless of opacity.
• Noise bleeding where it shouldn't - wire a mask_optional; the node multiplies the final result through the mask region only.

One honest caveat: this is image-space noise, so it lands before the sampler in a classic pipeline - it's a preprocessing/side tool, not a replacement for the sampler's latent noise. Use it where you can see it, and it does exactly what it says.