Wildcard Encode (Inspire)

This node does three jobs at once: it expands wildcards and dynamic prompts, it loads any LoRAs you name inline in the text, and it encodes the whole thing into conditioning - using BlenderNeko's advanced CLIP encoder rather than ComfyUI's default. It's the Impact Pack's ImpactWildcardEncode, rebuilt to encode with CLIP Text Encode (Advanced) so you get A1111-style prompt weighting on top of the wildcard machinery.

If you've come from A1111 and miss typing __hairstyles__ to roll a random option, or <lora:detail:0.6> right in the prompt, this is the node that brings both to ComfyUI in one place. It's especially handy because it hands back a patched model and clip and the encoded conditioning, so a single node covers "roll the prompt, apply the LoRAs, encode it" instead of a chain of three.

How it works

You write your prompt in wildcard_text using the Impact/dynamic-prompt syntax: __name__ pulls a random line from a wildcard file, {a|b|c} picks one of the options, and <lora:name:weight> loads a LoRA inline. On run, it populates the random choices into populated_text, loads the referenced LoRAs onto the model/clip, and encodes the result. The mode controls the randomness: populate re-rolls every run, fixed keeps what's there, and reproduce restores a specific past populated_text so you can recreate an earlier result.

The inputs and outputs that matter

• wildcard_text - your prompt with wildcards, {a|b|c} choices, and inline <lora:…> tags.
• mode - populate (reroll), fixed (keep), or reproduce (restore an exact past populate).
• weight_interpretation - how prompt weights are read: comfy, A1111, compel, comfy++ (the default), or down_weight. Pick A1111 if you're matching weights from an A1111 prompt.

Outputs: model and clip (patched with the inline LoRAs), conditioning (the encoded prompt, straight into a KSampler), and populated_text (what the wildcards actually rolled).

Installing it

ComfyUI Manager: search "Inspire Pack", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Inspire-Pack

Then restart. This node has two hard dependencies the README spells out: the Impact Pack (for the wildcard processing) and BlenderNeko's Advanced CLIP Text Encode (ComfyUI_ADV_CLIP_emb, for the encoding). Install both.

Where people get tripped up

Miss either dependency and the node either won't load or won't encode the way you expect - the Impact Pack supplies the wildcard engine, the Advanced CLIP encoder supplies the weighting, and Wildcard Encode is just the seam between them. Second thing: wildcards only work if you actually have wildcard files. The __name__ syntax reads from text files in the wildcards folder, so an empty setup means __hairstyles__ just sits in the prompt as literal text. And if you're chasing reproducibility, remember populate re-rolls on every run - switch to fixed or reproduce once you've got output you like, or your "final" render won't match the one you were happy with.