🔎 Egregora Region Select

Regional prompting has a workflow problem: once Egregora Region Plan has split your image into nine pieces, how do you know which piece is region 3 without squinting at a grid? Egregora Region Select is the answer - it's a tiny inspection node that pulls one region out of the list so you can actually look at it while you write its prompt.

What it does

You give it three things: the region_plan (from Egregora Region Plan), the regions_list (the per-region IMAGE list that same node outputs), and a region_index. It returns that single region as an IMAGE. That's the whole node. Region numbering is row-major, so region 1 is the top-left, region 2 is the one to its right, and so on down the rows - the same order the region plan's region_order_text describes and the same order Egregora Regional Conditioning expects prompts.

If you hand it an index out of range it doesn't error - it clamps to the nearest valid region, which is handy when you're sweeping through indices with a widget while captioning. It's a viewer, not a transformation; there's nothing about the output you need to worry about corrupting your data.

Why you'd actually use it

Three real jobs:

• Manual prompt writing. Pull up region 4, write "a nice white ferrari on a grassy field" for region 4. Seeing the actual crop beats reasoning from a grid every time.
• Captioning / VLM loops. If you're feeding regions to a captioner, this is the node that hands the captioner one crop at a time instead of trying to make sense of a whole batch.
• Debugging region order. When the plan produces more regions than you expected, or you're unsure whether a region is a core crop or a context crop, this lets you spot-check. The regions_list from Region Plan is the core crops (the tight per-region boxes), so what you see here is the part each prompt owns - useful for deciding whether a prompt boundary is in a sensible place.

The honest take

This is a utility node, not a magic one. If you don't care about inspecting regions before writing prompts - say, because a captioner is doing all the work and you're not curating its output - you can skip it entirely and the workflow still runs. The pack works fine with Regional Conditioning fed straight from a captioner. Reach for Region Select the moment you're hand-authoring prompts, because guessing what a region contains from coordinates is miserable and this removes that guess entirely.

Installing

It ships with the rest of the pack, so:

cd ComfyUI/custom_nodes
git clone https://github.com/lucasgattas/ComfyUI-Egregora-ARMD

then restart, or install "ComfyUI-Egregora-ARMD" via ComfyUI Manager. No extra models, no downloads - the pack's dependencies (numpy, opencv-python, PyWavelets, torch) are already in a stock ComfyUI environment. Like the rest of the Egregora family from Lucas Gattas (mrgattax), it's straightforward research-grade tooling: the node itself does exactly one thing, and it does it without surprises.