XY Input: LoRA Block Weight

Block weighting a LoRA is powerful, but the honest problem is you never know which blocks to touch. This is the node that answers that. It feeds the Efficiency Nodes' XY Plot with a set of block-weight vectors so you can generate a whole grid - same prompt, same seed, one row per vector - and see at a glance which blocks change the image and which do nothing. It turns "guess a vector, regenerate, squint, repeat" into one render you read off a wall.

If you've ever wanted just the style of a character LoRA without its pose, or you're trying to understand why a LoRA fights your composition, this is how you actually learn the layout instead of copying someone else's presets on faith.

How it works

You give it a list of block vectors in block_vectors - the default is a stack of preset pairs like SD-INS/SD-ALL, SD-MIDD/SD-ALL, SD-OUTS/SD-ALL, each comparing one region of blocks against the full LoRA. It outputs two XY axes, X (vectors) and Y (effect_compares), which you wire into an Efficiency Nodes XY Plot. When that runs, you get the grid.

The interesting extra is xyplot_mode: Simple just renders each vector, Diff shows the difference from the baseline, and Diff+Heatmap overlays a heatmap so the regions that changed light up. The heatmap_palette (viridis, magma, and friends), heatmap_alpha, and heatmap_strength control that overlay. It's a genuinely nice way to see a LoRA's structure rather than reason about it.

The inputs and outputs that matter

• lora_name - the LoRA you're profiling.
• block_vectors - the multiline list of vectors/pairs to sweep. This is the real control.
• xyplot_mode - Simple, Diff, or Diff+Heatmap; the heatmap mode is why you'd use this over a plain XY setup.

Outputs are X (vectors) and Y (effect_compares) - both feed the Efficiency Nodes XY Plot node.

Installing it

Get the Inspire Pack via ComfyUI Manager (search "Inspire Pack", install, restart), or manually:

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

Restart afterward. The requirements.txt matters more here than for most Inspire nodes - the heatmap rendering leans on matplotlib, which is in there, so let Manager install it (or run pip install -r requirements.txt yourself).

Where people get tripped up

This node does nothing on its own. It's an input for someone else's plot: you need the Efficiency Nodes pack (jags111's fork) installed, and the pack's README notes you want a build from after September 3rd for block-weight support. The X and Y outputs only make sense wired into that pack's XY Plot with the dependencies connected - miss that and you'll have a node that produces two mystery outputs and no grid. Set the Efficiency side up first, then hang this off it.