Doom Random Prompt

Writer's block is a real workflow problem, and Doom Random Prompt is the pack's cure: it assembles a prompt from category word-lists shipped in the pack, with two wildcard syntaxes you can use to shape exactly how random "random" gets. Set a seed and it becomes reproducible - the same seed, the same counter, the same prompt, every time. That's the difference between a dice roll and a workflow you can actually debug.

How it works

The pack ships 13 category files in its lists/ folder - background, characters, clothing, color_palette, composition, decorative, gender, lighting, mood, motion, special, style, techniques. The node reads those, picks one line from each enabled category, and stitches them into a prompt. Each category has an enable_<name> toggle, so you can, say, lock out style and techniques when you only want scene descriptions.

Two wildcard syntaxes extend it beyond the fixed categories:

• {a|b|c} - pick one of the alternatives inline.
• __filename__ - pull a random line from lists/filename.txt at that spot in the prompt.

So you can write a template prompt where half the words are fixed and half are rolls, which is the sweet spot: structured enough to be coherent, random enough to keep surprising you.

The inputs that matter

• counter - increments each run, and it's part of what makes a seed reproducible. Feed the counter output back into this input and it climbs by one per run.
• seed - -1 means a fresh random pick every run. Any non-negative value makes the result reproducible: same seed + same counter = same prompt.
• enable_* - the category toggles. All default on.
• seed_used (output) - the seed that was actually used. Log this; it's what lets you recreate a prompt you liked three weeks later.

Outputs are random_prompt (the assembled string), counter (input + 1, for the loopback), and seed_used.

Setup and the catches

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/PeterMikhai/Doom_Flux_NodePack

Restart ComfyUI, or search "DoomAI Nodes" in Manager. No dependencies.

Three things are worth knowing before you lean on it:

• The lists are yours to edit. They live in the pack's lists/ directory. The shipped categories are a start, not a contract - the real power is replacing the lines with your own vocabulary. Add your own .txt and the node picks it up as a new category on restart.
• "Reproducible" requires discipline. Same seed and same counter. If the counter drifts (say you changed the loopback wiring), the "same" seed gives a different prompt. That's by design - it's what makes counter an intentional knob.
• It's a starting point, not a finished prompt. These categories produce describe-the-scene text, which suits Flux and Krea 2 fine. On strict SDXL-style tag models, you may want to pipe the result through a prompt builder or edit it down.

One honest note: the shipped vocabulary is English and generic - you're not going to get "the exact niche aesthetic from that one artist" out of the box. But as a slot machine for "give me something I didn't think of", it's exactly the right amount of machine, and the seed logic is done properly, which is rarer than it should be.