⭐ Star Dynamic LoRA

The stock ComfyUI LoraLoader handles one LoRA, which means stacking three of them means three nodes daisy-chained, and stacking ten means a graph that looks like a plate of spaghetti. Star Dynamic LoRA is the StarNodes answer: one node with dynamically appearing slots. Fill the last slot and a new one appears; each slot has a name, a strength, and an on/off toggle. It's the LoRA-loading pattern from the pack's "Stack multiple LoRAs" story, and it's one of the most genuinely useful nodes in StarNodes.

It's in ⭐StarNodes/Sampler - the same category as the pack's unified sampler it typically feeds.

How it works

The model input is required. The clip input is optional - and this is the design decision that matters: leave CLIP disconnected and LoRAs are applied to the model only; connect it and each LoRA's single strength value is applied to both the model and the CLIP. One strength for both, which is the sane default for most LoRAs (the alternative - separate model/CLIP strengths per slot - is what makes LoRA loaders confusing).

Each slot is three inputs: loraN_name (a dropdown of your installed LoRAs, default None), strengthN (default 1, range -100 to 100 - negative strengths work, which is how you do anti-LoRA effects), and enabledN (a boolean toggle). The schema only shows lora1_name because the extra slots are dynamic - they appear automatically as you populate the last one. In the source, the node reads every loraN_name it finds in its kwargs and applies them in slot order, so the stacking order is just slot order.

The two outputs are model and clip, ready to feed a sampler. None slots are skipped cleanly, so half-filled slots are harmless.

Why you'd use it

Any workflow that stacks multiple LoRAs - style LoRA plus character LoRA plus a detail booster is the classic trio, and the on/off toggles make A/B testing one LoRA in a stack trivial without deleting nodes. The model-only variant (see Star Dynamic LoRA (Model Only)) exists because some models - particularly Flux-family and newer transformer models - want LoRAs applied to the model without touching the CLIP/text-encoder, and the clip handling can get finicky there.

Installing it

Standard StarNodes install - ComfyUI Manager, search Starnodes, install, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/Starnodes2024/ComfyUI_StarNodes
cd ComfyUI_StarNodes
pip install -r requirements.txt

Search the canvas for star - it's under ⭐StarNodes/Sampler.

Gotchas

Two things trip people up. First, dynamic slots mean the graph is slightly non-deterministic in what it shows - if a saved workflow references more slots than the node currently displays, the node still applies them because it reads by index, but the visual layout may look incomplete until you touch it. Second, the single strength applies to both model and CLIP when CLIP is connected; if you have a LoRA that wants a strong model weight but a weak CLIP weight, this node won't give you that separation - use it when the unified strength is fine, which is most of the time, and reach for a finer-grained loader for the edge cases. The negative-strength range is a nice bonus most beginners don't know exists.