CLIPTextEncodeWildcards2

CLIPTextEncodeWildcards2 is the sibling of this pack's main wildcard encoder, and the difference is one input and a whole lot of reproducibility. Same job - expand a wildcard prompt, encode it, hand the sampler CONDITIONING - but it adds a seed field, and the node calls random.seed(seed) before drawing from the wildcard files. Same seed plus the same prompt gives you the same expansion, every time.

That's the thing to reach for when you're not just batch-farming variety but trying to debug it. You roll a great draw on generation 17, and with the plain no-seed node you're stuck recreating it by luck. With this one you note the seed, and the identical combination comes back. It also dodges the pack's most annoying quirk: because ComfyUI caches unchanged inputs, the seedless variant won't re-roll unless you edit the prompt text. Here, you just tick the seed up a notch to force a fresh draw.

The inputs and output

Three inputs, all straightforward:

• text (multiline) - your prompt with __wildcard__ references and {a|b|c} choice syntax.
• clip - from your checkpoint or a CLIP loader.
• seed (INT) - pins the random draw. Bump it to re-roll.

One output: CONDITIONING, wired into a sampler's positive or negative slot. The underlying engine reads its card files from the wildcards folder at your ComfyUI root (the level above custom_nodes); each .txt file contributes lines as options, __name__ grabs a random line, and {n$$a|b|c} picks several.

Install

Same story as the rest of the pack - clone and restart. ComfyUI Manager finds it if you search "Lilly":

cd ComfyUI/custom_nodes
git clone https://github.com/lilly1987/ComfyUI_node_Lilly

Restart after cloning. No models to download; the pack auto-installs rich and chardet on first launch and prints colored debug lines to the console.

One honest caveat: this whole pack is unmaintained - the author's README says as much and recommends Impact Pack's wildcard node instead. For a 2023-era SD 1.5 or SDXL workflow that's already wired around this pack, the seed input here is genuinely the most useful way to make wildcard batch generation reproducible. For a new project, pick the maintained alternative and save yourself the archaeology.