SD1 MBW Layers Binary

The float version of this node (SD1 MBW Layers) lets you blend each UNet block by any weight from 0 to 1. This one strips that down to a hard switch: every field is a BOOLEAN, default false. Check a block on and, in the merge it feeds, that block comes fully from the other model; leave it off and it stays fully from the first. No blending, no in-between - which is exactly the point. The pack's own README describes these binary variants as making it "easy to extract specific layers," and that's the honest use case: this is a diagnostic tool for isolating exactly which block a trait lives in, not a blending tool.

The layout

Same 25-block shape as the float SD1 MBW Layers node, just booleans instead of floats:

• input_blocks.0 through input_blocks.11 (12 booleans)
• middle_blocks (1 boolean)
• output_blocks.0 through output_blocks.11 (12 booleans)

All default false - meaning nothing's flipped, which behaves like alpha 0 for every block rather than a neutral merge. You have to deliberately check boxes for this node to do anything.

Output

A single MBW_LAYERS, same as every other MBW node here - wires into SDNext Merge's optional_mbw_layers_alpha.

Why bother with binary instead of just using the float node

Because 25 fractional sliders is a huge search space to explore blindly, and this node lets you narrow it fast. Flip roughly half the blocks on, merge, and check whether the trait you're chasing shows up. It does - narrow to that half and repeat; it doesn't - try the other half. A few rounds of that bisection gets you from "somewhere in 25 blocks" down to one or two specific candidates, at which point the float SD1 MBW Layers node is the right tool to dial in exactly how much of that block's contribution you actually want, rather than the all-or-nothing this node gives you.

Installing it

ComfyUI Manager: search TechNodes, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/TechnoByteJS/ComfyUI-TechNodes --depth 1

No models to fetch, no dependencies to install.

Where people get burned

The most common misuse is treating this like the float node and expecting gradation - it doesn't have any, that's not a bug. If you want a partial blend, switch to SD1 MBW Layers instead. Second, the real value of this node comes from discipline: flip on several blocks at once on your first attempt and you can't tell which one actually caused whatever changed. Toggle one block at a time, merge, generate, compare against your baseline, and only then move to the next block - that's what "extracting specific layers" actually means in practice. And as with the float version, this is the SD 1.5-shaped node (12/1/12 blocks) - pair it with SD 1.5 checkpoints in your merge, not SDXL ones, or the block count won't line up.