Lora Selector v2

If you want to run more than one LoRA at a time on your SDXL generation, this is the Searge node for it. It gives you five LoRA slots, each with its own strength, all in a single node - no daisy-chaining five separate LoRA loaders. It feeds the selections into the workflow's data stream, and the workflow applies them to the model and CLIP for you.

It's under Searge/UI/Inputs, so it's a current v4.x node (the "v2" in its display name), and multi-LoRA was one of the headline features when Searge rewrote the pack for v4: up to five LoRAs stacked at once. That was a nice convenience in the SDXL era, where mixing a character LoRA with a style LoRA and a detail LoRA is completely normal.

How it works

Each of the five slots is a dropdown of the LoRA files you have installed, paired with a strength value. Pick a file, set how hard it applies, repeat for as many slots as you need. Leave a slot on none (or its strength at 0) and it's simply skipped. The node bundles all of this into the SRG_DATA_STREAM, and the workflow's model-loading stage reads it and applies each LoRA in turn.

One thing you'll notice on a fresh or cloud install: if no LoRA files are present in ComfyUI/models/loras, the dropdowns show only none. That's not the node being broken - it populates from whatever you've actually downloaded. Drop some .safetensors LoRAs into that folder, restart, and they'll appear.

The inputs and outputs that matter

• lora_1 … lora_5 - the file for each slot; none disables the slot.
• lora_1_strength … lora_5_strength - how strongly each applies. Range is −10 to 10, but you almost never want anything wild; 0.6–1.0 is the normal working band, and stacking several near 1.0 can quickly overcook an image.
• data in/out (SRG_DATA_STREAM) - the workflow spine that carries the selections onward.

Practical stacking advice

More LoRAs is not more better. Each one pulls the model toward its training, and three at full strength fight each other into mush. The usual approach: keep the LoRA that matters most near 1.0, and dial the supporting ones down (0.4–0.7) so they season rather than dominate. If a character stops looking like themselves once a style LoRA joins, that's the interaction - lower the style strength first. Negative strengths exist for pushing away from a concept, but that's an advanced move, not a default.

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: python -m pip install opencv-python in ComfyUI's Python env (required for the pack), then cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, restart. Windows portable has an installer script.

The node needs your LoRA files in ComfyUI/models/loras. SDXL LoRAs are portable across the SDXL finetune families, with the usual caveats between Pony and Illustrious lines.

Where people get burned

• Empty dropdowns. No LoRAs installed = only none shows. Add files to models/loras and restart.
• Everything at strength 1.0. Stacking five LoRAs at full power is the fast route to a fried image. Balance them.
• Base/finetune mismatch. A LoRA trained on a Pony checkpoint may misbehave on an Illustrious one. Match the LoRA to its intended base family.