SD1 MBW Layers

Merge Block Weighted (MBW) is actually older than SDXL - it started life as a dedicated A1111 extension (sdweb-merge-block-weighted-gui) built specifically for SD 1.5's UNet, letting people assign a different merge weight to each block instead of one flat alpha for the entire model. This node is that same idea, ported to ComfyUI: it doesn't merge anything by itself, it's a dial board that outputs one MBW_LAYERS bundle for SDNext Merge's optional_mbw_layers_alpha input to consume.

The layout

25 floats, all 0–1, all defaulting to 0.5 (which is identical to a flat, unweighted merge - you have to actually move sliders for this node to change anything):

• input_blocks.0 through input_blocks.11 - 12 encoder blocks.
• middle_blocks - the bottleneck.
• output_blocks.0 through output_blocks.11 - 12 decoder blocks.

That 12-in/1-mid/12-out shape is specific to SD 1.5's smaller UNet - SDXL's is 9-in/1-mid/9-out, which is why there's a separate SDXL MBW Layers node rather than one shared field set. Feed this SD1 node's output into a merge of two SDXL checkpoints and the block counts don't line up; use the right one for your base architecture.

SD 1.5 is still very much a going concern in 2026 despite everything that's shipped since - plenty of people still run it, still merge it, and MBW's original use case was exactly this: figuring out which of its 25 blocks a specific trait (a face, a style, a bad habit) actually lives in.

Output

A single MBW_LAYERS - wires into SDNext Merge's optional_mbw_layers_alpha, where it overrides both the flat alpha and any mbw_preset_alpha selection.

Installing it

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

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

Nothing else to install.

Where people get burned

Architecture mismatch is the main one - this node is SD 1.5-shaped, and using it against an SDXL merge (or the reverse, using the SDXL node against SD 1.5 checkpoints) simply doesn't fit. Beyond that, treat hand-tuning 25 sliders as genuinely exploratory: the original MBW community's own advice was to play with the presets first to get a feel for what each block does, because it's hit-or-miss without that intuition. SDNext Merge's mbw_preset_alpha dropdown ships 39 named presets for exactly this reason - reach for this node once you already know which block you're chasing, and consider the SD1 MBW Layers Binary sibling if you want a hard on/off per block to isolate a trait before dialing in how much of it you want.