ClybAdaptiveLoraLoaderModelOnly

ClybAdaptiveLoraLoaderModelOnly is the trim variant of the pack's LoRA stacker. Same adaptive UI, same up-to-20 slots, same clone-once-and-merge-all trick - but the CLIP input is gone, and the CLIP strength is forced to 0.0 for every slot. Only the diffusion-model patches get applied. It returns a single MODEL, nothing else.

Why would you want that? Mostly for the workflows where you don't have a CLIP to patch, or where patching it is actively wrong:

• Unconditional / low-guidance pipelines. At CFG 1 many modern flow models don't even compute the unconditional pass, and there's no text encoder in the loop to modify.
• Flows with a separate text encoder. If your CLIP (or T5) is wired in through its own path, you don't want a LoRA loader stomping on it.
• Model-only chains like img2img or refinement passes where you want a stylistic LoRA on the model and nothing touching conditioning.

The LoRAs still land in your models/loras folder, the strength_model_N widgets still go negative if you want to subtract a LoRA's effect, and the same dynamic UI applies: fill slot 1 with a real LoRA and slot 2's widgets appear. A slot with strength_model = 0 is skipped.

Wiring and install

Output is the patched MODEL, ready for your sampler:

model → ClybAdaptiveLoraLoaderModelOnly → KSampler(model=…)

Installation is identical to the rest of the pack - no dependencies:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/Clybius/ComfyUI-ClybsChromaNodes.git
# restart ComfyUI

Or grab it through ComfyUI Manager by searching "ComfyUI-ClybsChromaNodes".

The honest take

If you're doing normal text-to-image with a prompt, use the full ClybAdaptiveLoraLoader instead - it's a strict superset, and forgetting to load a clip-only LoRA patch is an easy silent mistake to make. This variant earns its keep in exactly the niche it names: model-only sampling. Same EXPERIMENTAL flag as the parent, so keep the pack updated when ComfyUI bumps core.