LoRA Picker

The Nynxz LoRA Picker is the same multi-LoRA stack widget as the pack's loaders, with one crucial difference: it has no model and no clip sockets. It's headless - you build the stack, and it just outputs the stack on a wire. Then separate Apply LoRA nodes use that stack wherever you need it.

Why the indirection

Because one picker can feed several Apply nodes. You get one place to build and tune a stack - the searchable picker, preview thumbnails, bookmarks, per-row toggles and strengths - and then fan it out to multiple checkpoints or multiple points in the graph. Change a strength once and every consumer updates. It's the "one source of truth" pattern from the plumbing layer, applied to LoRA stacks.

The wiring is: LoRA Picker (build the stack) → lora_stack wire → Nynxz Apply LoRA (model + clip in, model + clip out). The Picker's single input is the stack widget; its single output is lora_stack of type NYNXZ_EXP_LORA_STACK_DATA. That type is the pack's own, so you can't accidentally wire it into a stock node - it only mates with the pack's Apply LoRA. Bookmarks persist in favorites.json beside the pack, so your curated set survives restarts.

Loader or Picker?

The workflow question is when this beats the loaders. If you have exactly one checkpoint in the graph, the plain LoRA Loader is fewer nodes and does the same job. The Picker earns its keep the moment the same stack needs to reach two places - the classic case being a hires-fix pass where you want the same character stack applied at both the base and the upscale stage, or two model variants being A/B'd with the same LoRAs. If you never do that, you can skip it and not feel guilty.

The trade-off is real: the stack lives in one node, so the graph reads as "here's a magic bundle going to the Apply node," and if you save a workflow the picker's stack state is what travels - which is a feature for sharing (one definition, no per-node drift) and a mild trap for the "wait, which stack was this again?" moment. Name your nodes.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Nynxz/ComfyUI-NynxzExperimental

or search "Nynxz" in ComfyUI Manager, then restart. No dependencies or model downloads; your LoRAs stay where they are. And since this is the headless piece, remember it does nothing until an Apply LoRA node is connected - a lone Picker is just a widget waiting for work.