Wildcards

This is the node the pack leads with in its own README screenshots, and it's the closest thing Chibi-Nodes has to a signature feature. It swaps a placeholder keyword in your prompt for a random entry (or several) pulled from one of ten built-in word lists - animals, art-styles, clothes, colours, descriptors, fabrics, foods, household-objects, landscapes, poses - so instead of manually varying your prompt across batches, you let the node roll the dice.

If you've used the __wildcard__ convention from other tools (Dynamic Prompts and similar extensions scan a folder of arbitrary .txt files and let you reference any of them by filename), the idea here is the same in spirit but smaller in scope: Chibi's Wildcards node ships with exactly ten fixed categories baked into the pack itself. There's no folder to drop your own lists into and no nested __file__ syntax - you pick one of the ten from a dropdown per node. If you need a custom word list or want to combine multiple categories in one prompt, you'll want a dedicated wildcard-processor node (this pack itself, and others, ship separate "wildcard processor" style nodes for that) rather than expecting this one to scan arbitrary files.

How it works

You choose a textfile from the dropdown, tell it what keyword to look for in your prompt text (defaults to the literal string __wildcard__), and how many entries_returned to pull - 1 to 10. Feed it your prompt via the optional text input, and it replaces occurrences of the keyword with randomly chosen lines from the selected list. Give it a seed and the picks become reproducible; leave it and every queue rerolls.

The inputs and outputs that matter

• textfile - which of the ten bundled lists to draw from.
• keyword - the placeholder string to find-and-replace in your prompt; change it from the default if __wildcard__ collides with something else in your text.
• entries_returned - how many random lines to pull and insert, 1 through 10.
• seed (optional) - fix it for a reproducible pick, randomize it for variety.
• clip (optional) - connect a CLIP model and the node will encode the result straight into a CONDITIONING for you, skipping a separate CLIPTextEncode.

Two outputs: CONDITIONING, populated only if you fed a clip; and text, the finished string with the keyword swapped out, useful even without a CLIP connection if you just want to see (or further process) the resulting prompt.

How to install it

• ComfyUI Manager: search "ComfyUI-Chibi-Nodes", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/chibiace/ComfyUI-Chibi-Nodes/, restart.

The word lists ship inside the repo - no separate download, no extra Python dependencies. That's worth calling out because a lot of wildcard tooling in the wider ecosystem expects you to source your own text files; this one works the moment you install it.

Common issues

The most common miss is forgetting the keyword has to appear literally in your text input - if you type __wildcard__ in one place but change the node's keyword field elsewhere (or vice versa), nothing gets replaced and your prompt goes out unchanged with the placeholder text sitting in it verbatim. Double-check the two match.

The other trap is scope: because the ten lists are fixed and general-purpose (colours, foods, animals), don't expect this node to give you domain-specific variety - art-style names, camera terms, or anything niche to your subject. It's built for broad, general randomization, not a curated prompt-engineering wordlist. If you outgrow the ten categories, that's the signal to look at a full wildcard-processor node (Chibi's own pack includes related "wildcard/lora syntax processor" nodes, and dedicated wildcard extensions exist elsewhere) rather than fighting this one to do more than it's built for.