Conditional LoRA Loader

Most LoRA loaders in ComfyUI are all-or-nothing. Either the LoRA is applied or you're right-clicking the node to bypass it mid-workflow. Conditional LoRA Loader is the stock ComfyUI LoraLoader with exactly one widget added - enabled - and that small addition quietly fixes a workflow chore.

First, the name. "Conditional" here doesn't mean conditional models or conditioning. It means the loader has a switch. The node's own description reuses the generic ComfyUI LoRA text ("modify both diffusion and CLIP models, altering the way in which latents are denoised"), and the actual behavior is dead simple: toggle on and the LoRA applies; toggle off and your model and CLIP pass through untouched.

How it works

The source is a couple dozen lines. The class subclasses ComfyUI's built-in nodes.LoraLoader - the exact same loader behind the default "Load LoRA" node - and adds enabled to the required inputs. When enabled it calls the inherited load_lora; when disabled it just returns [model, clip]. Because it inherits rather than reimplements, everything you already know about the built-in loader is true here: the lora_name dropdown reads from models/loras, you can stack several of these in series, and negative strengths work exactly as they do on the vanilla loader.

Inputs and outputs

The inputs that matter:

• model and clip - feed in whatever your checkpoint or prior loader produced.
• lora_name - the dropdown of LoRAs in models/loras.
• strength_model (default 1, range −100 to 100) - how hard the LoRA hits the diffusion model. Negative is a real, occasionally brilliant trick for subtracting a baked-in style.
• strength_clip (default 1, range −100 to 100) - the CLIP-side weight, which matters for LoRAs that touch the text encoder.
• enabled (default true) - the whole point.

Outputs are MODEL and CLIP, wired exactly like any other loader.

Why you'd reach for it

The killer use case is A/B testing. Put your style LoRA on the chain, fix a seed, run with enabled on, flip it off, run again - same composition, no rewiring, no hunting for the bypass option in the right-click menu. Because enabled is a widget rather than a workflow hack, the on/off state travels with the saved workflow, so a friend (or future you) gets the comparison setup intact. It's the same idea behind rgthree's Power Lora Loader toggles, but scoped to a single LoRA and with zero extra machinery.

Installing it

The pack publishes to the Comfy Registry (its pyproject.toml carries the [tool.comfy] block), so the clean path is ComfyUI Manager → search "ComfyUI Simply Nodes" → install → restart. Manual clone works too:

cd ComfyUI/custom_nodes
git clone https://github.com/Wicloz/ComfyUI-Simply-Nodes

No requirements.txt, no model downloads. The only dependency is ComfyUI itself, since the loader leans on the built-in nodes.LoraLoader.

Common issues

• Toggle seems to do nothing: if the LoRA is also loaded elsewhere in the graph, or its strength is 0, the switch changes nothing visible. Check strength_model is non-zero.
• Weak or zero effect when on: this is LoRA behavior, not the node - missing trigger words are the #1 cause (the troubleshooting docs are full of it). Check the model's page for the trigger, and remember default 1.0 strength is often too strong, not too weak.
• Not seeing a LoRA in the dropdown: it needs to be a .safetensors (or .pt) in models/loras, and you may need to refresh the node to rescan.

If you only ever run one LoRA at full strength, the stock loader is fine and you don't need this. If you switch styles a lot and are tired of bypass-muting, this is the one to keep in the graph.