BKWILDCARDS Selector

If you've ever stared at a wildcard cheat sheet and thought "there has to be an easier way," this is the node that proves you right. BKWILDCARDS Selector doesn't ask you to type __wildcard__ tokens or learn {a|b|c} syntax. It ships roughly 4,800 hand-written entries across 50 categories and seven themes, and turns them into scoped dropdowns. Pick a theme, choose your categories, wire one output into your prompt, and generate.

The catch up front: this node only produces text. It draws one line from each enabled category and joins them into one prompt string, out of a single output called prompt (type STRING). Wire that into your positive prompt block - a CLIPTextEncode node's text input, or a string node like StringFunction that's already feeding your prompt. The node nags you if you forget: it turns red - NOT ready, please wire the node to your prompt block. - and green (ready - press Run) once connected.

How it actually works

Under the hood it's refreshingly honest. Each category is a text file with named sections; the dropdown choices map to those sections. On Run, the node draws one entry from each enabled category and joins them with a separator. The seed drives everything: each category gets its own RNG seeded from your seed plus a stable checksum offset, so the same seed always reproduces the same prompt across sessions and processes. That determinism is also why the on-node preview never lies: preview and generation call the same resolution function, so what you see at queue time is byte-identical to what the image is made from.

Scoping is enforced in Python, not the browser. Select _ghost.runner and only that theme's categories can contribute, no matter what the widgets say. The bundled JS extension just hides off-theme dropdowns; if it fails to load, the node still behaves correctly, it just shows everything.

The inputs that matter

There are a lot of widgets, but you actually set a handful. Theme is the seven bundled releases (_ghost.runner, Whimsical Woods, Autumnal Oxidation, Cassette Futurism, All the Dresses, Nettie Necket, plus an adult lingerie pack); only the selected theme's categories contribute. Gender is - off -, - random - (rolls per seed), Female, Male, or Fluid (both sets of physical options, an androgynous subject). Then the category dropdowns, most reading - off -, - random -, or a named section - off contributes nothing, random draws from the whole file. Small categories like face, nose, lips, and spell casting are simple on/off toggles. Under Settings you get seed (set its control to randomize for a fresh combo per queue), separator (default , ), label_output, and mayhem. label_output on gives you structured lines like gender: a woman, / hair: ... / scene/background: ...; plain gives one comma-joined string. mayhem is one-click chaos - it ignores every selection and composes a fully random cross-theme image. Fun, but it includes the adult pack, mind that when demoing.

A couple of smart touches: a Bald hair pick suppresses hair color and style, and the Cybernetics Color choice is folded into the augment's finish word rather than emitting separately. The resolved text is also embedded into generated PNGs via the workflow metadata, so dragging a finished image back into ComfyUI restores the exact prompt that made it.

Install

Via ComfyUI Manager (search BKWILDCARDS) or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/bkidderz/BKWILDCARDS.git

Then restart ComfyUI. There are no dependencies - Python standard library only, no model files to download. The node title shows the build number (e.g. BKWILDCARDS Selector 0.9.10) so you can confirm an update took.

Common issues

• Output box frozen but new [BKWILDCARDS] node … lines appear in the server log each run - the backend is fine and the frontend script is stale; restart ComfyUI or hard-refresh the browser.
• Toggles for other themes are visible - the JS extension failed to load. It fails open; behavior is still correct because theme gating lives in Python.
• Dropped in your own wildcard file and it doesn't appear - the category scan is cached at import. Restart, or use Refresh Node Definitions.
• Prompt text not in your saved PNGs - the embed rides ComfyUI's metadata path, so anything that strips PNG text chunks (re-encodes, screenshots, most hosts) loses it. Same fight every ComfyUI workflow has, not a BKWILDCARDS bug.

License note, since it matters if you redistribute: the software is MIT, but the wildcard content is CC BY-SA 4.0 - share-alike applies to the library, not the images you generate. Those are yours.