SEGS Wildcard Prompt

Wildcards - the {a|b|c} dynamic-prompt syntax that Impact Pack made famous - are usually a per-image thing: the whole image picks one option and you move on. XJSegsWildcardPrompt does something more interesting: it parses wildcard syntax per segment, so segment 0 gets one prompt, segment 1 gets the next, and so on. It's the bridge between "detector found N things" and "each of those things gets its own tailored prompt."

The node understands a genuinely useful slice of Impact Pack's wildcard grammar. The headline feature is [SEP]: you write your prompts separated by [SEP] markers, and the node hands back the prompt matching your index, cycling if the index exceeds the count. That's how you map segment index → prompt. On top of that:

• Ordering modes - [ASC], [DSC], [ASC-SIZE], [DSC-SIZE] control prompt order, and [RND] shuffles (seeded by the seed input, so it's reproducible).
• [SEP:R] and [SEP:SEED] - assign a random or fixed seed to a specific prompt's {a|b|c} expansion, so the same option list gives different results per segment.
• [LAB] mode - map prompts to labels: [LAB][face]promptA[hand]promptB[ALL]shared. Pair it with the label input (feed it the label output of SEGS Extractor) and segments get prompts by what they are, not just by position.
• Option expansion - {a|b|c} basics, weighted {3::a|b}, and multi-select {2$$,$$a|b|c}.

The interface:

• wildcard_text - the multiline wildcard string.
• index - the segment index (0-based).
• label - optional; required for [LAB] mode.
• seed - optional; drives [RND] shuffling and random expansions.
• Output: prompt, the resolved STRING for that segment.

Where it earns its place: the per-region detail workflow. Loop segments through this node (index = the loop counter), get a per-face or per-object prompt, and feed it to the detailer's positive conditioning. That's the difference between detailing every face with the same generic prompt and giving the main subject "portrait of a woman, detailed eyes" while the person in the back gets "candid, motion blur." The [LAB] mode is the more robust route if your detector's labels are reliable - it survives segment order changing between runs.

Install is the pack-wide routine:

cd ComfyUI/custom_nodes
git clone https://github.com/alexjx/ComfyUI-XJNodes

Restart ComfyUI, find it under XJNodes/segs, or use ComfyUI Manager and search "ComfyUI-XJNodes". No models, empty requirements.txt. The grammar it parses comes from Impact Pack's wildcard world, so familiarity with that syntax pays off here.

Caveat: this is a personal-use pack with a thin footprint, and the node implements a subset of Impact Pack's grammar - don't assume every wildcard feature from other tools works here. What's implemented is documented in the source and covers the common cases well.