LoRA Loader Model Only Selector

LoRA Loader Model Only Selector is a normal LoRA loader wearing a switcher's hat. You stuff up to sixteen LoRA files into its slots, then drive a single index input to pick which one actually loads. Wire that index to a For Loop Start's index output and you've got a batch workflow that iterates through every LoRA in your list automatically - no editing the graph between runs. This is the "model-only" variant, so it patches just the diffusion model, not the text encoder; the full LoraLoaderSelector sibling also touches CLIP.

How it works

Mechanically it's a thin wrapper around ComfyUI's own comfy.sd.load_lora_for_models. On each run it collects all sixteen lora_0 through lora_15 slots (dropdowns populated from your models/loras folder), counts the non-"None" ones, clamps index into range, and loads loras[index] at your chosen strength. If the slot is "None" or strength is 0, it passes the model through untouched. The node also caches the last-loaded LoRA weights in memory, so re-running with the same file skips re-reading it from disk.

Inputs and outputs

• model (required) - the MODEL to patch
• index - 0 to 15, which LoRA to load (required)
• strength - −100 to 100, default 1.0 (required)
• lora_0 … lora_15 - the LoRA pickers; leave unused ones at "None"

Outputs are model (patched), lora_name (the filename that loaded, or "None"), and count (how many non-None slots you configured). count is quietly useful - you can clamp a loop to it so you never iterate past the last real LoRA.

The README mentions a bypass toggle, but the shipped code doesn't actually expose one - to skip all LoRA loading, set strength to 0 or clear the slots, and the model passes through unchanged. Use the Reset All button on the node (added by the pack's JS) to zero out every slot at once.

Install

Via ComfyUI Manager (search "Latent Astronaut Suite") or:

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

Restart. No extra dependencies - it uses ComfyUI's built-in LoRA machinery.

Where people get burned

The classic trap is forgetting that this version doesn't touch CLIP. If you're on a model where the LoRA's text-encoder side matters (some SD/SDXL style LoRAs depend on CLIP changes), the model-only version will apply it wrong or barely at all - reach for LoraLoaderSelector instead. Also remember the dropdowns only show files already in ComfyUI/models/loras, so a fresh LoRA needs a refresh before it appears. And keep in mind all sixteen slots share one index - a common bug is leaving old files in low slots and wondering why "index 2" loads something you forgot about.