Bubba Prompt Randomizer

Bubba Prompt Randomizer is a structured character-prompt generator hiding behind a node. Instead of one giant prompt field, you get a dropdown per category - hair color, hair style, outfit, pose, expression, setting, weather, body build, and a couple dozen more - and it assembles the final booru-tag prompt for you, pulling one value from each category you turn on. It's the machine behind "I want a random character, every run, that still fits my scene."

The whole thing is backed by JSON files shipped in the pack's data/prompt_randomizer folder, one file per category, each a list of tag choices (expression.json has everything from smile to seductive; there's a full 1.character_Presets set too). The node builds its dropdowns from those files at load time, which is why the category widgets in your UI have names like hair_color and setting - they're literally the JSON filenames. Want to add a category? Drop a new JSON file in that folder and restart.

How it works

Each category dropdown has three states: disabled (don't include it), random (pick one per run), or a specific tag you chose. A seed input drives all the picks, so the same seed reproduces the same character - the usual ComfyUI reproducibility contract. On top of the categories you get freeform fields: prefix_text (stuck at the front), extra_positive (appended), and negative_prompt (used as-is). cleanup and dedupe normalize and de-duplicate the assembled string, and remove_category_underscores strips the underscores from tags if you're targeting a model that prefers plain spaces. Wire in a pipe and a clip and you also get ready-to-sample conditioning.

The inputs and outputs that matter

For a beginner the three freeform text fields are where you'll spend your time; the two dozen category dropdowns are the fun part. Leave a category at disabled and it's ignored entirely - no need to hunt through them all.

Outputs to know:

• positive_prompt / negative_prompt - the final assembled strings. This is what gets saved into metadata and shown in the PNG.
• chosen_values - a per-category report of what was actually picked this run (like expression: smile, hair_color: brown). Invaluable when you love an image and want to know which combination produced it.
• positive / negative - conditioning, only when a CLIP is connected.

How to install it

Same pack as everything else here: ComfyUI Manager → "Bubba Nodes" → install → restart, or

cd ComfyUI/custom_nodes
git clone https://github.com/bubbafett5611/bubba_nodes.git

restart ComfyUI, pip install -r requirements.txt if needed (aiohttp and pydantic, both light). No models to download - the categories ship in the package. Needs ComfyUI v0.27.0+.

Gotchas

The seed here is a plain widget, so remember the control_after_generate timing trap from the node-plumbing docs: the seed shown is the one that runs next unless you set "widget control mode" to Before. And keep the tag reality in mind - the KB's prompt-engineering essay makes the point that these categories are only as good as the model's tag vocabulary. On a Pony or Illustrious checkpoint, a fashion_style tag like gothic_lolita is gold; on a Flux-style LLM-encoded model, a pile of booru tags wants to be a sentence. The randomizer assumes the tag world. Also note the pack is new with essentially no community tutorial trail yet - the JSON files in the data folder are self-documenting, so open them if you want to know exactly what you're picking from.