Load Flux LoRA

Of everything in this pack, this is the node most likely to be sitting unused in your workflow because you don't actually need it. Load Flux LoRA does exactly what its name says - apply a LoRA to a Flux model at a given strength - and it does that job fine, but it's largely a product of timing: it shipped alongside the rest of XLabs' Flux tooling in the days right after Flux's August 2024 launch, when solid native LoRA support for Flux in stock ComfyUI wasn't yet a given. These days ComfyUI's own LoraLoaderModelOnly handles Flux LoRAs natively without needing this pack at all. The honest reason to reach for this one specifically now is if you're already using other XLabs nodes and want your LoRA folder to live in the same xlabs/ convention, or if you're loading one of XLabs' own published LoRAs - they put out a handful themselves, RealismLora among them, alongside the ControlNets and IP-Adapter.

The mechanism, briefly, in case LoRA is new to you: a LoRA is a small file encoding a low-rank delta on top of the base model's weights - it doesn't replace the checkpoint, it nudges it toward a style, subject, or concept. This node takes your Flux model, applies that delta scaled by strength_model, and hands back a modified MODEL ready for the next node in your chain.

The three inputs: model is your Flux checkpoint (this loader is Flux-specific - a LoRA trained for SDXL or SD1.5 will not work through it, LoRA files are locked to the architecture they were trained on). lora_name picks the file, populated from whatever's sitting in your LoRA folder. strength_model sets how strongly it applies, with a default of 1 - the range technically runs from -100 to 100, which is far wider than you'll ever practically use; realistic values sit close to 0–1.5, with negative numbers there mainly if you deliberately want to subtract a LoRA's effect rather than add it. The output is a plain MODEL, wire it straight into your sampler or chain another LoRA loader after it.

Installing it: search "x-flux-comfyui" in ComfyUI Manager, or manually cd ComfyUI/custom_nodes && git clone https://github.com/XLabs-AI/x-flux-comfyui and restart (the manual path also wants a one-time python setup.py run from inside the folder). This node doesn't need its own separate download - but for lora_name to show any options at all, your .safetensors LoRA files need to be sitting in ComfyUI/models/xlabs/loras, which the pack creates automatically on first launch.

The one thing worth flagging: if you're training your own Flux LoRAs to load here rather than downloading someone else's, Flux training practice has moved a fair distance since this pack's era - rank/alpha and learning-rate defaults that worked for SDXL don't transfer, and modern training guides are worth reading before you assume old SDXL numbers apply. That's a training-time concern rather than anything about this loader node, though - whatever .safetensors file you end up with, this node will load it the same way regardless of what trained it.