LoRA Loader (CLIP)
The multi-LoRA stack loader when you also need to patch CLIP
- model
- clip
- stack
- model
- clip
Some LoRAs patch more than the UNet - they patch the text encoder too, and for those you need the model and CLIP to pass through the same stack so both get the same treatment. That's the whole reason this variant exists. LoRA Loader (CLIP) takes MODEL and CLIP in, runs them through the on-node multi-LoRA stack, and gives both back out, patched together.
The stack, with a CLIP knob
The stack widget is identical to the plain LoRA Loader's, and it's the point of the pack's whole LoRA section: rows, each with an on/off toggle, a searchable picker with preview thumbnails, and a strength - all on the node instead of a daisy chain of single-LoRA loaders. Bookmarks persist in a favorites.json beside the pack. The one extra this variant adds: per-row strength applies to both model and CLIP, and you can set a separate CLIP strength per row if a LoRA needs a lighter touch on the encoder.
Inputs and outputs are the simple part: model + clip in, the stack widget, model + clip out, wired straight into your positive/negative CLIP encode and the sampler. Stock ComfyUI users will recognize this as the multi-stack analog of LoraLoader - the classic SDXL-era pattern where style LoRAs patch CLIP L and character LoRAs lean on the UNet.
When you actually need CLIP patching
Mostly the older SD1.5/SDXL line - and the modern checkpoints that don't need it (Flux, Krea, Qwen-based) are why the pack ships a MODEL-only loader as the default. If your workflow runs fine without CLIP patched, the plain loader is one less wire to think about. If you're on an SDXL character or style stack, this is the one - and it's the right shape for the anime/illustrious-style combos where the CLIP side genuinely moves the result.
The honest caveats are shared across the pack: it's experimental, so schemas and widget behavior can shift between commits, and there's no pack-level dependency or model download - your LoRAs live in your existing models/loras folder. The stack row strengths apply to model and CLIP together by default; remember the per-row CLIP override exists, because the default of "same strength everywhere" is exactly the thing people forget to check when a stack that should be subtle comes out shouting.
Installing it
cd ComfyUI/custom_nodes
git clone https://github.com/Nynxz/ComfyUI-NynxzExperimental
or search "Nynxz" in ComfyUI Manager, then restart. Keep ComfyUI reasonably current - these nodes use the newer V3 node API.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| clip | CLIP | — | |
| stack | NYNXZ_EXP_LORA_STACK | LoRA stack — add rows, pick + bookmark on the node |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| clip | CLIP | — |