SeedCluster

If you've ever wanted "the same scene, but a few different takes" without hand-typing eight seeds into eight samplers, this is the node. UC_SeedCluster takes one master seed plus an increment and emits a cluster of eight related seeds: the master, then seed+increment, seed+2×increment, and so on, wrapping safely inside ComfyUI's 64-bit seed range. One knob to turn, eight seeds to play with.

How it works

The schema is refreshingly small. seed (default 0, full 64-bit range, with control_after_generate - so you can set it to randomize and roll the whole family at once) and increment (default 1) go in. Two outputs come out: seed as a plain INT (so the master can drive a sampler directly), and a Cluster of type UC_SEED_CLUSTER - a custom type holding all eight values.

The cluster output isn't meant to be wired into a sampler by itself. It feeds this pack's companion UC_FromSeedCluster node, which pulls individual seeds out of the cluster by index. That's the intended flow: SeedCluster → FromSeedCluster → KSampler, with a small "index" widget controlling which of the eight variants you're sampling right now. Bump the index, re-run, and you're stepping through a coherent family of related results instead of jumping to unrelated random seeds. Neighboring seeds in most diffusion models give you variations on the same composition - close enough to compare prompts or LoRA weights, different enough to actually differ.

When it's worth it

Grids and comparisons. Test a LoRA at eight seeds, compare two samplers across the same cluster, or do an "8 variants of this prompt" batch where you want the set to be comparable. The increment input is the interesting knob: 1 gives you eight consecutive seeds (maximally similar), while a large increment spreads them out (more diversity while still being derived from one master). For a quick sanity check of how much a change matters, a cluster with increment 1 is the honest baseline.

Install

Part of ComfyUI-UtilsCollection:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-UtilsCollection

Restart, or use ComfyUI Manager (search "UtilsCollection"). No model downloads; the pack's requirements (opencv-python, typing-extensions) auto-install.

Notes

Watch the wrap: at the top of the seed range the cluster wraps within 0xFFFFFFFFFFFFFFFF, which is fine - ComfyUI's own seed range does the same. And the after-vs-before widget timing applies to the master seed's randomize control; if you want the box to show the seed that actually ran, set widget control mode to Before in settings. Minor pack quirk: UtilsCollection registers some primitive/logic IDs that other packs also claim, so expect the occasional "replace node?" prompt when loading workflows built on the older standalone packs.