Time to Seed List

Random seeds that change on their own

Every ComfyUI beginner learns it the hard way: set a seed, render, and the image is stuck. To vary output you have to go poke the KSampler's randomize button - and in a scheduled or unattended workflow there's nobody around to poke it. Time to Seed List is this pack's answer: a one-trick utility that hands you a list of random seeds, re-rolled on every single execution.

Let's be honest about what it is: a tiny node, one input, one output, no settings beyond a number. But it's the difference between an overnight batch that replays the same image four hundred times and one that produces four hundred variations.

How it works

It reads the current time (HH:MM:SS), uses it to seed Python's random number generator, and spits out count random integers across the full 32-bit seed range (0–4294967295). The clever bit is IS_CHANGED: it returns a fresh timestamp every call, so ComfyUI's caching never kicks in and the node re-executes each run. A normal node only re-runs when its inputs change; this one deliberately invalidates itself so the seeds are never stale.

The one input, and the honest wiring note

• count - how many seeds you want (1–100000, default 1).

The output is seed_list, an INT list - and that's where beginners trip. A KSampler wants a single INT seed, not a list. To use all N seeds you need something that iterates the list (a loop or iterate node from another pack); if you only want one fresh seed per run, grab a single element and feed that.

Which is also the most honest use case here: the pack's scheduler fires the same workflow at 6am, this node hands you a fresh seed or two, and today's render isn't a clone of yesterday's. For one-off interactive runs you might as well just hit randomize on the KSampler - this node earns its keep in unattended mode.

Gotchas

• Seeds derive from seconds, so two runs inside the same second produce identical seeds. If you want one set per day rather than per second, the pack's Daily Prompt Scheduler does date-seeded selection - different tool, different job.
• If generation fails internally, the node falls back to returning a list of 42s instead of erroring. An easy trap: a run of identical "random" images is a symptom, not a coincidence.
• No models, no heavy dependencies - it's part of the same ComfyUI-ScheduledTask pack as the scheduler, nothing extra to download.

Install

• ComfyUI Manager: search "ComfyUI-ScheduledTask", install, restart.
• Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/dseditor/ComfyUI-ScheduledTask
cd ComfyUI-ScheduledTask
pip install -r requirements.txt

Only schedule and requests get installed - nothing GPU-heavy. You'll find the node under utils.