List Custom Seed

Seed management is where a lot of ComfyUI workflows quietly rot. You want reproducible results, so you don't want fully random seeds every run - but when you're batch-testing prompts or styles, you also don't want to hand-type ten seed values. This node splits the difference: give it one seed and a count, and it deterministically generates that many unique seeds, as a list.

It was added way back in the pack's v1.2.7 era with the phrase "control after generate," which is the whole personality of the thing. Because it's seeded from your input, the same seed + count always produces the same list - your test run is reproducible. But you can also just bump the seed (or re-run the graph) to get a fresh set, which is "control after generate" in practice.

How it works

Two inputs, two outputs.

• seed - your master seed (default 42). This becomes the seed for Python's random, so the output list is a deterministic function of it.
• count - how many seeds you want (default 3, up to 1000).

It draws count random integers in the full seed range, skipping any duplicates it happens to hit, until it has count unique values. Outputs are seed_list (a list-type output of INTs) and count (the actual number generated, useful for loop logic). The seeds are clamped to the valid range, so nothing comes back negative or out-of-bounds.

The typical wiring: seed_list into a per-item processing pattern (each iteration pulls one seed from the list), with count feeding the loop's "how many iterations" input. Every iteration gets a distinct, reproducible seed, and the whole batch is pinned to your one master seed.

Installing it

It's part of 1hewNodes. ComfyUI Manager → search "ComfyUI 1hewNodes" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/1hew/ComfyUI-1hewNodes

Restart ComfyUI. No special dependencies, no model downloads.

Common issues

The reproducibility contract is the thing to protect: the list is deterministic only if the node runs the same way every time. If something upstream forces a re-eval, you get the same list again (that's the point), but remember the list itself is random-looking - seed 42 and seed 43 give entirely unrelated seed lists, so "just set it to 42" won't reproduce a specific prior batch unless 42 was the master seed you used. Also, seed_list is a list type, not a batch - convert it if your downstream expects a tensor. And one practical note: because it's Python's random, not a GPU RNG, don't expect the seeds to correlate with anything in ComfyUI's own seed generator - they're just valid numbers for feeding back in.