LoRA Loader Selector

LoRA Loader Selector is the full-strength version of the pack's LoRA switcher: sixteen LoRA slots, one index input to pick which one loads, and - unlike the model-only sibling - it patches both the diffusion model and the CLIP text encoder. That CLIP side is the whole reason to pick this one. Style and character LoRAs on SD/SDXL often bake their identity into the text-encoder weights too, so if a LoRA looks like it's not taking effect with a model-only loader, this is the node you want. It also plays perfectly with the pack's For Loop Start, letting you sweep through LoRAs automatically.

How it works

Same machinery as the model-only variant, plus CLIP. Each run it collects the sixteen lora_0 through lora_15 dropdowns (populated from your models/loras folder), counts the non-"None" ones, clamps index into 0–15, and loads loras[index] through ComfyUI's comfy.sd.load_lora_for_models with separate strength_model and strength_clip weights. If the slot is "None" or both strengths are zero, model and CLIP pass through untouched. The loaded LoRA's weights are cached in memory, so re-running with the same file skips the disk read.

Inputs and outputs

• model (required) - the MODEL to patch
• clip (required) - the CLIP to patch
• index - 0 to 15, which LoRA loads
• strength_model, strength_clip - independent strengths, default 1.0 each. Dropping strength_clip to ~0.7 is a common starting point for style LoRAs.

Outputs: model and clip (both patched), plus lora_name (the loaded file, or "None") and count (how many non-None slots are configured - handy for clamping a loop so it never runs past your real LoRAs).

Install

ComfyUI Manager, search "Latent Astronaut Suite," or:

cd ComfyUI/custom_nodes
git clone https://github.com/latentastronaut/comfyui-latent-astronaut-suite

Restart. Uses ComfyUI's built-in LoRA loader, so no extra dependencies.

Where people get burned

The mirror-image trap of the model-only node: this one always touches CLIP, so if you use it on a model whose text encoder doesn't accept the LoRA's CLIP weights (common with some newer or video architectures), you can get errors or subtly degraded conditioning. If in doubt, check whether your LoRA is distributed as "model only" - those don't have a CLIP half and should go through the model-only loader. The Reset All button on the node clears all sixteen slots at once; otherwise, an old LoRA hiding in a low slot is the usual "why is it loading the wrong file" culprit.