Load LoRA Block Weight

This one's for a specific, slightly power-user corner of the ecosystem: LoRA block weighting. If you've never wanted to apply a LoRA at different strengths to different layers of the UNet, you can safely skip this node. If you have - and it's a real, if niche, technique with a steady trickle of community use - Load LoRA Block Weight is how you reload a preset you saved earlier.

Block weighting comes from hako-mikan's sd-webui-lora-block-weight, and the Inspire Pack brings the idea to ComfyUI. The premise: a LoRA doesn't affect all UNet blocks equally, and by dialing individual blocks up or down you can isolate what a LoRA does - keep a style but drop the character it drags along, or stop a LoRA from wrecking your composition while keeping its texture. It's surgical, and the payoff is real when a LoRA is "almost right but."

How it works

The Inspire block-weight nodes form a little chain. Make LoRA Block Weight builds an LBW_MODEL - a LoRA with a per-block weight vector applied - instead of baking it straight into your model. Save LoRA Block Weight writes that out as a .lbw.safetensors file. And this node, Load LoRA Block Weight, reads one of those files back into an LBW_MODEL so you can reuse a preset you tuned once. From there you'd apply it to your MODEL and CLIP with the Apply node.

So it's the "load a saved recipe" step. The value is not re-deriving a block vector you already got right - save it once, load it whenever.

The inputs and outputs

• lbw_model - a dropdown of your saved .lbw.safetensors files. Pick the preset to load.
• Output is a single LBW_MODEL, which you feed into Apply LoRA Block Weight to actually apply it.

That's it - it's a loader, so there's not much to it beyond picking the file.

How to install it

ComfyUI Manager: Install Custom Nodes → search ComfyUI Inspire Pack → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Inspire-Pack

Then restart. Inspire Pack is Dr.Lt.Data's work, so it installs cleanly. No model downloads for the node itself - but see below about the files it reads.

Common issues

The number-one confusion: the dropdown is empty on a fresh install. That's expected - there are no .lbw.safetensors files until you've made some. Load can only see what Save has written, so the workflow is Make → Save first, then Load in later runs. If you dropped in a workflow that uses this node and the dropdown is blank, you're missing the saved preset file the author had locally; you'll need to rebuild it with Make/Save or get the .lbw.safetensors from them. And remember Load only produces the LBW_MODEL object - it doesn't apply anything by itself; you still need the Apply LoRA Block Weight node downstream, or nothing happens to your generation.