🎲 Bawk Wildcard Encoder

The conditioning node of the pack, and the one doing the most work. Bawk Wildcard Encoder takes your prompt, resolves any {wildcard|syntax}, applies up to six LoRAs to your model and CLIP, and encodes the result into FLUX conditioning - with the resolved prompt passed out the side so your image saver can log what actually generated the image. In a stock workflow that's a text encode node, two LoRA loaders, and a few reroutes. Here it's one node.

How it works

Three steps in sequence. First, if wildcard_seed is not -1, it runs the prompt through _process_wildcards: {sunset|night city|rainy street} picks one option at random, seeded so a fixed seed gives a fixed choice. Set wildcard_seed to -1 and wildcard processing is off - the braces stay literal in your prompt, which is how you disable it. Second, the six LoRA slots: each has lora_N_on, lora_N_name, lora_N_strength, and a slot only loads when the toggle is on, the name isn't "None", and the strength isn't 0. Finally it tokenizes and encodes with the LoRA-patched CLIP (encode_from_tokens with pooled output - the FLUX way), and the conditioning comes out the CONDITIONING socket.

The inputs that matter

• prompt - your text with {a|b|c} wildcard syntax. This is old A1111 wildcard syntax that a lot of people who arrived after 2024 never learned, but it's exactly as powerful as it ever was.
• wildcard_seed - controls which random option wins. -1 disables wildcards entirely.
• The LoRA slots - lora_1_on/name/strength through lora_6_on/name/strength. The README's suggested usage: slot 1 for the main character/style at 0.8–1.2, and smaller strengths (0.2–0.6) as you go down for fine detail. Strengths go -10 to 10, so negative values subtract a LoRA's influence.

Outputs and wiring

MODEL and CLIP come out LoRA-patched (the pack chains them back through so the sampler sees them), CONDITIONING feeds Bawk Sampler, and PROMPT_OUT is the fully-resolved prompt - wire it to Bawk Image Saver's processed_prompt and you'll always know which random variant made which image. That's genuinely the killer feature for batch work.

Install

Same pack install - ComfyUI Manager → search "Bawk Nodes" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/juddisjudd/ComfyUI-BawkNodes.git

No model downloads beyond your LoRAs in models/loras/. requests (in the pack's requirements) is there for the AI enhancement feature.

One honest caveat

The README sells AI prompt enhancement - send your prompt to Claude, GPT, or Grok and get a richer version back. Here's the thing: in the shipped v2.3.0 code, the API-call functions exist but no UI inputs are registered for them. There's no toggle, no API-key field, nothing wired up. The marketing is ahead of the code. Don't go hunting for the settings - they're not there yet. The wildcards and LoRA slots work exactly as described, so you lose nothing you'd actually use daily.