PD: Random Prompt

Say you're generating a sticker pack and want a dozen playground items in the same art style - swing, slide, sandbox, merry-go-round. You could hand-type forty prompts, or you could let this node build the combinations and feed them straight into a batch. That's PD: Random Prompt: a list of things, a fixed template, and it cross-products them into ready-to-encode prompts, with an option to shuffle or just take the first N.

The default template ships with the exact use case baked in: the mutable list is a pile of playground equipment, and the immutable prompt is sticker, Cartoon, `` - the backtick placeholder is where each item gets dropped in. It's a template engine with one placeholder, and that's most of what you need for batch variation.

How it works

It reads mutable_prompt line by line (one item per line, blank lines stripped) and treats immutable_prompt as a template. Any line containing `` gets the placeholder replaced with each mutable item; a template without the placeholder gets , `` appended so the item lands at the end. Every mutable item × every template line becomes one prompt. Then:

• random_sample = enable shuffles with random.sample; disable takes them in order.
• max_count caps the output (default 9, up to 1000).
• If you set seed to anything nonzero, it seeds Python's random so the same run is reproducible. Seed 0 means "don't bother" - you'll get a different shuffle every run.

The output is a list of strings (the node declares OUTPUT_IS_LIST), so it's meant to be fed into something list-aware downstream - a batch image saver, a queue, or a node that iterates prompts. A progress bar shows up during generation, which is a nice touch from the pack.

The inputs that matter

• mutable_prompt - your variable items, one per line. The default playground list is a decent demo; you'll replace it.
• immutable_prompt - the fixed template; use `` as the item placeholder.
• random_sample - enable / disable.
• max_count (1–1000) - cap on how many prompts come out.
• seed (optional) - lock it for reproducibility.

Installing it

Part of Comfyui_PDuse by 7BEII. ComfyUI Manager (search "PDuse") or:

cd ComfyUI/custom_nodes
git clone https://github.com/7BEII/Comfyui_PDuse.git
cd Comfyui_PDuse
pip install -r requirements.txt

Restart. No models to download - the pack's deps are numpy, Pillow, scipy, opencv-python, torchvision. Chinese-first docs (Feishu/Bilibili).

Where people get burned

The biggest gotcha: the output is a list, and a regular CLIP Text Encode node doesn't take lists. You need a list-aware encoder or a batch/queue system downstream, or you'll get a type error and a confused afternoon. Second, shuffling happens once per run over the built combinations, not per line - two runs with the same seed give the same prompts, but "same seed, different run" only holds if you actually set one. Third, max_count caps but doesn't pad: if you ask for 9 and only built 4 combos, you get 4. If you want every combination, set random_sample = disable and max_count high.