⭐ Star Dynamic LoRA (Model Only)

Some models don't want their text encoder touched when you apply a LoRA. Flux-family and the newer transformer-based models are the ones people hit this on most - the LoRA is meant to patch the diffusion model, and feeding the CLIP into a LoraLoader that insists on dual application is either a no-op or a headache. Star Dynamic LoRA (Model Only) is the StarNodes variant that drops the CLIP entirely: same dynamic slots, same toggles, but the only thing being modified is the model.

It's in ⭐StarNodes/Sampler, right next to its sibling Star Dynamic LoRA.

How it works

The schema tells the whole story. One required input, model. Then the dynamic slots - lora1_name (dropdown of your installed LoRAs, default None), strength1 (default 1, -100 to 100), enabled1 (boolean) - and as you fill the last slot, more appear. Each slot applies its LoRA to the model at that strength, in slot order. The single output is model, ready for a sampler.

That's it. No clip input exists, no clip gets passed through, nothing to leave disconnected by accident. It's the version of the dynamic LoRA loader you reach for when you've already decided the CLIP stays out of it - or when your graph simply doesn't have a CLIP wire running through this part of the pipeline (increasingly common with Qwen/Z-Image-style setups where the text encoder lives elsewhere or the model handles conditioning internally).

When to pick this over the regular version

If your workflow already has a CLIP line and you want LoRAs applied to both model and CLIP, use Star Dynamic LoRA and connect it. If your model is one of the transformer-generation ones where LoRAs are model-only by convention, or if there's no CLIP wire to route anyway, this node keeps the graph honest - no phantom CLIP connection, no "did I forget to connect it?" moment. For Qwen-Image and Z-Image-Turbo workflows specifically, model-only LoRA application is the pattern most LoRA makers assume, so this is often the correct pick out of the box.

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

The same dynamic-slot caveats as the sibling apply: the graph may display only the always-present first slot until you interact with it, and the slots still apply by index even if the visual layout looks partial. And the obvious one - there's no CLIP handling here, so if you do need a LoRA's text-side effects, this node won't deliver them. For most model-only LoRAs that's a non-issue; for dual-purpose LoRAs with meaningful CLIP weights, use the regular node instead. Don't blame the tool for the one case it explicitly doesn't cover.