🔀 Bawk LoRA Switch

This is the node you reach for when you've got more than one character, style, or subject set up and you're tired of bypassing nodes and re-wiring your graph to switch between them. Bawk LoRA Switch takes up to four LoRA stacks - the kind produced by the pack's Bawk LoRA Stack node - and passes exactly one through, selected by a single A/B/C/D dropdown. It's the A/B test lever for the whole Bawk Nodes workflow: flip it and you're generating a different character with zero rewiring.

How it works

The mechanism is almost insultingly simple, which is why it's reliable. Four inputs (set_a through set_d), one selector, one output. The node looks at the active widget, grabs whichever set you pointed at, and forwards it as a LORA_STACK. It does not merge, blend, or sum - it's a routing valve, the "A/B switch" shape from ComfyUI's plumbing layer, not a fallback. And like the Bawk LoRA Stack, it doesn't load weights itself; the pack's Wildcard Encoder is what actually applies the LoRAs to the model. "Only the chosen set is loaded" is shorthand for "only the chosen set's data reaches the encoder," which in practice is the same thing.

The part people underestimate is that trigger words travel with the set. Each set you built with a LoRA Stack carries its own trigger words, and the encoder injects them into the prompt only while that set is active. So switching from set A to set B doesn't just swap LoRAs - it swaps the prompt vocabulary too. You type the character's name once in the Stack's trigger field and never think about it again.

The inputs and outputs that matter

• active - the dropdown. A to D, default A. This is the only thing you'll touch day to day.
• set_a … set_d - each takes a LORA_STACK, almost always the output of a Bawk LoRA Stack. Leave one unwired and the switch just passes an empty stack for that position, which is harmless.

Outputs match the Stack's, because they feed the same places:

• lora_stack - into the Wildcard Encoder's lora_stack input.
• lora_triggers - the active set's trigger words as one string. Useful to know you're actually getting the words you expect.
• lora_info - the name@strength, … summary for the pack's image saver, so the PNG metadata records which set was live.

Installing it

Same pack, same install. Bawk LoRA Switch ships in juddisjudd's Bawk Nodes Collection, so you get the whole FLUX workflow in one shot:

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

Restart ComfyUI after cloning, or use ComfyUI Manager and search "Bawk Nodes". The pack's only dependency is requests, but it wants a ComfyUI build from late 2025 or newer (FLUX.2 support), and ComfyUI-GGUF if you want GGUF models and LoRAs in the dropdowns. The repo's flux1_dev_lora_sets.json example shows two stacks feeding a switch feeding the encoder - copy that wiring.

Gotchas

• It picks one, full stop. If you want sets combined - a character plus a style, say - don't route them both through a switch. Chain them instead: wire one stack into another stack's lora_stack input. The switch is for exclusive-or choices.
• The dropdown is a widget, not a socket. If you want to drive the active set from another node - say, to step through sets across a batch - right-click and convert active to an input, the same mechanic as any ComfyUI widget.
• It only shines inside the Bawk ecosystem. The stack type is the pack's own, so wire it into Bawk's encoder, not a stock LoraLoader. On its own, disconnected from a LoRA Stack and encoder, this node does nothing but pass an empty structure around.
• v3.0 of the pack reshuffled all Bawk widget layouts, so workflows saved against the old 2.x nodes need their Bawk nodes re-added after you upgrade.