π± PROJECT SEED
A scratchpad that remembers which seeds you've already tried
- seed
- history
Seed farming is the oldest habit in this hobby: run a bunch of seeds at low steps, find the one that's almost right, then re-run that exact seed properly. The workflow is great; the bookkeeping is the annoying part. Project Seed is a small node from the Artha-Projekt pack (by GitHub user Cyrostar) that keeps a running log of every seed a given run used, so you don't have to remember which of the forty numbers actually gave you that pose.
Honest framing first: it doesn't generate or randomize anything. It's a helper - a per-node scratchpad, plus a button that brings back the last seed you actually queued.
How it works
You give it a seed (a plain INT, default 42, up to the full 64-bit range) and it hands it right back unchanged on the seed output - wire that into your KSampler. Meanwhile it records the seed against a timestamp in a dictionary that lives on the node, and builds the history output: a string with one line per seed, like:
TIME 14:03:22 SEED 42
TIME 14:07:41 SEED 1793
TIME 14:11:09 SEED 8842
Wire history into a text-display or note node, or just read it off the node, and you've got your whole seed-run log in one place. On the frontend there's also a USE LAST SEED button: every time you queue a prompt, the extension stashes the seed value that was sitting in the widget, and the button restores it. That's the killer feature - you tweak the prompt, get garbage, and instead of scrolling your history to find the last good number, you click and it's back.
Where it earns its keep
It pairs naturally with the seed-farming trick from the concepts docs: DPM++ 2M at 10β15 steps to hunt seeds cheaply, then the full sampler/steps pass on the winner. Drop Project Seed into that low-steps branch, and the history string becomes your log of what you already tried - no separate notes file, no screenshot of your browser history.
The traps
The history is in-memory and per-node. It resets on ComfyUI restart, and each copy of the node keeps its own log - it won't aggregate across the graph. The USE LAST SEED button is also frontend state: it remembers what was in the widget when you last queued, not some server-side truth. It's a convenience, not a database.
And because it's a pure passthrough with no randomness, it's on you to feed it new seeds - from a randomizer, a fixed list, or by just typing them. The node remembers; it doesn't explore.
One more ComfyUI nicety for this one: it only depends on the standard library, so it's a genuinely zero-risk install.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/Cyrostar/Artha-Projekt
Restart, or grab it through ComfyUI Manager by searching "Artha-Projekt". No dependencies, no model downloads, no keys. If you already run the pack for Project Prefix, it's already installed - just drop the node in.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| seed | INT | 420β18446744073709550000 | β |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| seed | INT | β |
| history | STRING | β |