LoRA Block Info

Block weighting a LoRA - turning some of its layers up and others down instead of applying it at one flat strength - is one of those tricks that sounds fiddly until you see what it fixes. A character LoRA bleeding its background into every image? Often it's a few blocks carrying that style, and you can dial just those down while keeping the face. But to do that intelligently you need to know what's in the LoRA: which blocks exist, which ones are actually populated. That's what this node is for. It's the inspection step before you write a block-weight vector.

It doesn't change your image. It reads a LoRA and reports its block structure so you can make informed decisions in Inspire's LoRA Loader (Block Weight), whose whole syntax is a per-block vector. Think of Block Info as the map you consult before you start turning dials.

How it works

This is an output node - it produces a text report rather than a MODEL you wire onward. You give it a LoRA plus the model and CLIP it's meant to apply to, and it inspects the LoRA's weights against that architecture, listing the blocks the LoRA affects. The Inspire Pack's block-weight system is built on the same idea as hako-mikan's well-known sd-webui-lora-block-weight, where a block vector (using numbers, plus letters like R, A, B for randomized and preset values) controls each block's strength. Block Info tells you what those slots correspond to for this specific LoRA, so your vector isn't a blind guess.

The inputs that matter

• model (MODEL) and clip (CLIP) - the base you'll apply the LoRA to. The report is generated relative to this architecture, so use the same base you actually generate with.
• lora_name - pick the LoRA to inspect from the dropdown. This is the input you're really setting.
• block_info (multiline text) - the report field. The block breakdown gets surfaced here after you run.

There are no output slots - it's a terminal display node. Its job is to show you information, not pass data to the next node.

Installing it

Comes with the Inspire Pack by Dr.Lt.Data, the author of ComfyUI Manager and the Impact Pack, so it's core-ecosystem tooling. Via ComfyUI Manager: search ComfyUI Inspire Pack, install, restart. Manually:

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

then restart. No extra dependencies or downloads - but this node only earns its keep alongside the pack's LoRA Loader (Block Weight), which is where you'll actually apply what you learn here.

Common issues

"Where's the output? Nothing connects." By design. This is a read-only inspection node with no data outputs - it reports into its own text field. If you were expecting a MODEL/CLIP out to chain into a sampler, you want LoRA Loader (Block Weight) instead; Block Info just informs how you configure that.

The block names don't match another tool's. ComfyUI and A1111-style tools don't always label blocks identically. Use the report from this node when writing vectors for Inspire's block-weight loader - mixing conventions between ecosystems is how people end up muting the wrong blocks.

I don't know what to do with the info. The workflow is: inspect here, then in LoRA Loader (Block Weight) write a vector that lowers the blocks carrying the effect you don't want (often style/background) while keeping the ones carrying what you do (subject/face). It's iterative - read, adjust, generate, repeat.

Nodes won't load after install. Check the terminal log; the recurring Inspire Pack cause is a version mismatch. Update Inspire Pack and Impact Pack together and restart.