Seed List Generator (C2C)

Every batch run hits the same wall: you want ten images, all different, all reproducible, and the only seed source you have is a single random number you type into KSampler and hope. SeedListMEC is the fix - it takes one base_seed and hands you a full list of deterministic seeds you can wire into the graph. It's from the ComfyUI-CustomNodePacks umbrella (the big MEC pack), and it's about as close to a pure helper as that pack ships: no model, no VRAM, nothing to download.

Why you want this specifically: reproducibility. Consecutive seeds (base_seed, base_seed+1, …) give you images that are visibly related - great when you want a coherent set, frustrating when you're hunting variety. That's what the mode dropdown is for, and it's the whole ballgame here.

How it works

The node generates N seeds from a base seed three ways, matching the author's own description:

• increment - just base_seed+i. Fast, predictable, and the images will often feel like siblings.
• hash - sha256 of base_seed-index, first four bytes as a 32-bit seed. The point is de-correlation: the outputs scatter across the full seed space instead of creeping up by one, so they look unrelated while staying deterministic.
• random - a Python mt19937 stream seeded with base_seed. Also reproducible (same base, same list), but the mapping to your base seed is opaque.

All seeds are masked to 0xFFFFFFFF so they stay in the range a sampler expects.

The inputs and outputs that matter

Three inputs, two outputs - that's the whole node:

• base_seed (INT, default 0) - the seed everything derives from.
• count (INT, default 4, up to 1024) - how many seeds you want.
• mode (increment / hash / random) - the generator, described above.

Outputs:

• first_seed (INT) - the base seed itself, handy if you want the first sample to match what you already had.
• csv_all_seeds (STRING) - the full list as comma-separated text, e.g. 42,43,44,45. Wire it to a text preview to eyeball your sweep, or parse it downstream.

There's no batch-mode that auto-applies one seed per image - this node hands you the list and you decide how to consume it. That's a deliberate simplicity, and it's why the node doesn't bite.

Where it fits

The honest use case is batch sweeps with a fixed plan: base seed 42, five images, mode=random, and you know that the same graph will reproduce those five images tomorrow and next week. Pair it with the pack's TextTemplateMEC or FolderIncrementer if you're building per-shot prompt-and-save pipelines - seed list in, versioned files out.

Installing it

It's one node in the ComfyUI-CustomNodePacks pack, so install the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Code2Collapse/ComfyUI-CustomNodePacks.git

…then restart ComfyUI. ComfyUI Manager users can just search "CustomNodePacks" and click install. The pack's requirements (opencv-python, scipy, safetensors) are for its mask/VAE machinery, not this node - don't install anything extra for it, and don't blindly pip install -r requirements.txt because that can clobber ComfyUI's own torch/numpy. Look for [MEC] Loading MaskEditControl node pack … in the console to confirm.

The one thing to remember

If you want different-looking images, hash and random beat increment every time - consecutive seeds are the reason your "sweep" sometimes looks like five copies of the same picture. Everything else about this node is exactly as boring as it should be.