Generate Noise (Texturaizer)

The one node in this pack that isn't about reading Blender's export - it's a noise generator, meant to feed ComfyUI's advanced sampling chain (a NOISE object, the kind SamplerCustomAdvanced and friends expect, rather than the plain integer seed a basic KSampler takes). What makes it worth a dedicated node instead of the stock one is a dropdown most people don't think about until their seeds stop reproducing: which device generates the noise.

How it works, and why the noise device matters

Diffusion sampling starts from random noise, and how that noise gets generated - not just the seed number - affects the exact pixels you get. Generate noise on the CPU and generate it on the GPU with the same seed and you can get genuinely different starting noise, because the two code paths don't produce bit-identical random sequences. This node's own dropdown spells out the practical stakes directly: it labels one option GPU(=A1111), meaning if you're trying to match output from Automatic1111 (or reproduce a seed someone shared from there), that's the mode to reach for. The other option is a plain CPU path. If a seed you know from elsewhere isn't reproducing the image you expect, this setting - not the seed number - is usually the first thing to check.

Separately, batch_seed_mode controls how the seed advances across a batch of more than one image: incremental bumps the seed by a fixed step for each image in the batch, while comfy follows ComfyUI's own internal batch-seeding behavior. If you're used to how a stock ComfyUI batch spreads seeds and want that exact behavior reproduced here, comfy is the one to pick; incremental gives you a simpler, predictable step-by-step progression instead.

The inputs and outputs that matter

• noise_seed (0 to roughly 1.8 × 10¹⁹, i.e. the full 64-bit range) - your seed, same idea as any other seed field.
• noise_mode - GPU(=A1111) or CPU, per above.
• batch_seed_mode - incremental or comfy, per above.

Output is a single NOISE object - wire it into a SamplerCustomAdvanced node's noise input alongside your guider and sampler/scheduler pair, not directly into a basic KSampler, which takes a plain seed integer instead.

Installing it

ComfyUI Manager → search Texturaizer → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/LatentSpaceDirective/ComfyUI-Texturaizer

then restart ComfyUI. No extra dependencies - it's pure noise generation, nothing to download.

Where people get burned

Two traps, both about seed mismatch rather than anything broken. First: wiring NOISE output into a node that expects a plain integer seed (or vice versa) just won't connect - this is specifically for the SamplerCustomAdvanced-style advanced sampling chain, not the seed widget on a standard KSampler. Second, and more subtle: if you're trying to reproduce a specific render - your own from a previous session, or one someone else shared - matching the seed number alone isn't enough. noise_mode and batch_seed_mode both have to match too, or you'll get a different image from the "same" seed and spend time debugging the wrong thing.