h4 : Seed Broadcaster (v1.1.0) - Seed Generator

If you've ever tried to hold a seed steady across two samplers - a base pass and a refiner, say - you know the pain: both KSamplers have independent seed widgets, they drift apart, and suddenly your "controlled experiment" is testing two random variables at once. The community's first rule of seed discipline is "lock the seed, change one thing at a time," and this node exists to make locking actually possible. h4 : Seed Broadcaster (v1.1.0) - Seed Generator is one integer out, infinite fan-out: wire it into every sampler that needs to stay in sync and they all march to the same number.

It's a tiny utility in a tiny, shelved pack (zero search impressions - this is enthusiast territory), but it does its one job cleanly, and it shares the same install and support story as the rest of the h4 ToolKit.

How it works

Five widgets, one output. The important ones:

• mode - fixed holds the anchor seed, increment walks upward by increment_step each time, random rolls a fresh number using random_digits (1–12) of digits.
• auto_advance - when on (default), the node advances itself every execution. Turn it off to hold the current seed and approve changes manually.
• seed - the anchor. In fixed mode it's used directly; in other modes it's the starting point.

The output is a single seed INT. ComfyUI lets you multi-wire one integer to as many samplers as you like, so "one output, unlimited fan-out" isn't a marketing line - it's literally how the graph works.

One detail worth knowing: in increment mode the node logs every emitted seed even when the UI anchor sits at the start value, and changing modes or editing the anchor resets the internal pointer. So the sequence is reproducible, which is the whole point. There's also a Randomize button beside the seed field (added by the pack's h4_ui.js) for when you're in auto_advance = False and want a fresh roll on demand.

Install

Same as the rest of the pack - ComfyUI Manager, search "h4 ToolKit", or:

cd ComfyUI/custom_nodes
git clone https://github.com/m3rr/h4_ToolKit

Restart. No models, no heavy deps (just colorama and numpy, auto-installed at startup).

Where it fits

Pair it with The Engine when you're building grids - anchor every cell to the same seed for img2img refinement, or feed the Broadcaster's output to several samplers that must stay in lockstep. The README suggests teaming it with the Execution Logger to keep a seed history alongside tensor metadata, which is exactly the kind of audit trail that saves you an hour later when you can't figure out which run produced which image. It won't change what you generate; it just stops the seed roulette from sabotaging it.