Simple Prompt Library

Simple_Prompt_Library is the no-files version of the pack's flagship. Where Prompt_Library reads your prompts from .txt files on disk, this node keeps the whole library in a single multiline text box inside the workflow itself. Pick a prompt by index or let it randomize, and out comes the prompt plus the seed that produced the pick.

Why would you choose the text-box version over the file version? One word: portability. A file-based library only works on machines that have the file. A text-box library travels inside the workflow JSON itself - share the workflow, share the prompts, done. If you're building something you'll hand to other people (or run on a different machine), that's a genuine advantage. If your library is big enough to be a real database, you want the file version instead.

How it works

The prompt_text field uses the same syntax as the file version, so it should feel familiar:

• An empty line separates one prompt from the next.
• Lines starting with ### are titles - kept as structural placeholders, stripped from the output.
• --- or ---- splits positive from negative, mid-line allowed.

The default content demonstrates it:

### Masterpiece
masterpiece, best quality --- low quality, blurry

The random_seed input is where this node gets weird, and it's worth reading carefully because it does double duty as both an index and a seed:

• -1 → random pick every run (time-based), and the used_seed output stays -1.
• 0 or 1 → always the first prompt.
• 2–999 → treated as a 1-based index if it's within range; if it's past the end of your list, it falls back to acting as a random seed.
• 1000 and up → always treated as a random seed (reproducible for a given value).

Inputs and outputs

• prompt_text - your library. The only required text.
• random_seed - the index-or-seed hybrid above.

Outputs:

• prompt - the selected prompt as a string, --- separator intact.
• used_seed - an INT reporting which seed path the node took. Handy for logging or feeding into a seed node if you want to reconstruct a draw.

How to install it

Install the pack via ComfyUI Manager (search ComfyUI-Prompt_Library) or:

cd ComfyUI/custom_nodes
git clone https://github.com/hexxacubic/ComfyUI-Prompt_Library

Restart ComfyUI. Pure standard library, no pip step, no models to download. Category: hexxacubic.

Where people get burned

The seed semantics are the trap - nobody reads them on the first pass. If you set random_seed to 5 expecting "the fifth prompt" and your list only has three entries, the node doesn't clamp to 3; it treats 5 as a seed and picks something that looks arbitrary. Keep the value inside 1..list_length for deterministic indexing, or go ≥1000 for reproducible randomness.

Also, used_seed is not the actual seed used for the random draw - it's mostly an echo of your input (stays -1 on time-based picks). If you want to reproduce a good roll, the practical move is to write down the seed you fed it, not to trust the output to hand you a magic number. Same debugging rule that applies across this whole pack: fix the seed while you're tuning, randomize when you're hunting for surprises.