Wildcard Encode (divided)

Wildcard Encode is the pack's older, simpler wildcard node, and it deliberately does half the job. Where the newer Wildcard Divide rolls a prompt, splits it into regions, sizes your latent and hands you ready conditioning all in one, this node only does the rolling and the encoding - then it hands you a list of conditionings, one per [SEP] segment, and leaves the actual region work to its sibling, Comfy Divide. If you've seen this node's display name - "divided" - that's what it's pointing at: the output is already divided, just not spatially.

Mechanically it's straightforward. You feed it model, clip, and a wildcard_text containing __slot__ references and [SEP] separators. With mode set to Populate, it expands the wildcards on every run into populated_text (flip to Fixed and it just uses what's already in that box - handy when you want to stop rerolling). Each [SEP]-separated chunk gets CLIP-encoded independently, and the results come out of positives as a list: segment zero, segment one, and so on. The populated_text output shows you exactly what got rolled, which is how you debug a weird pick. The two dropdowns inject <lora:...> tags or __wildcard__ refs into the text field - pure UI convenience.

One genuinely nice thing buried here: the wildcard engine strips <lora:name:1.0> tags from the text, loads the LoRA into the model as part of the pass, and encodes the cleaned text. That means your wildcard files can carry their own style LoRAs instead of you hand-wiring LoraLoaders for each roll. It even supports LBW= block-weight syntax if you have the Inspire Pack installed; without it, the block weights are quietly ignored. LoRAs are applied cumulatively across all segments, so they're global - per-region LoRAs aren't a thing in this pack.

The one input that matters more than it looks: seed. It controls the wildcard roll, so the same seed reproduces the same pick. It's also wired so it can read from an ImpactInt or rgthree seed node rather than a bare integer, which matters if you're trying to reproduce someone else's workflow.

Where this fits: it's the middle step of the older two-node pipeline - Wildcard Encode rolls and encodes, Comfy Divide takes the positives list and does the spatial masking. If you already have Wildcard Divide, you don't need this node for anything except the list-based workflow or the LoRA-in-wildcard behavior. It has no negative input and no size handling; the division node owns those.

Install is the standard pack install - ComfyUI Manager, search "Wild Divide," or clone https://github.com/Julian-adv/WildDivide into custom_nodes and restart. No models, no extra dependencies. Same caveat as its siblings: the upstream repo has been returning 404 since mid-2026, so if the clone fails, that's the reason - find a mirror while it's still around.