FluxInspireLbw_Batch

LoRA Block Weight (LBW) is an old idea, going back to an A1111 extension for SD 1.5: instead of applying a LoRA at one global strength, you apply it per-block, because different blocks of the network tend to control different things - one might carry composition, another style, another fine detail. Set the right block strengths and you can keep the pose a LoRA taught without dragging its color palette along too. The "Inspire" in this node's name nods to ltdrdata's Inspire Pack, which built the best-known ComfyUI implementation of the same idea for SD 1.5/SDXL's block layout - this SP-Nodes version brings the concept to Flux instead, mapped onto Flux's own transformer structure of 19 double-stream blocks (which process image and text jointly) and 38 single-stream blocks (which process the combined stream afterward), rather than the old UNet's IN/MID/OUT layout.

How it works

This node doesn't apply a LoRA itself - it generates block-weight configurations that a compatible block-aware LoRA loader elsewhere in your graph then uses. And "Batch" means exactly that: it produces a whole set of different configurations in one go, built for sweeping through which blocks actually matter for a given LoRA rather than committing to one guess.

• block_strength (0–20, default 2.5) - the base strength applied to a block when it's enabled.
• double_blocks / single_blocks (booleans) - the coarse switches: turn LoRA application on or off for the entire double-block group or single-block group at once.
• force_double_blocks / force_single_blocks - comma-separated block indices that override the coarse toggles for specific blocks. The node's own tooltip gives the exact format: 1,5,6,8.
• forced_block_strength (0–20, default 1) - a separate strength dial that applies specifically to whatever you listed in the force fields, independent of block_strength.
• reverse_mode - flips the selection, so instead of "apply to these blocks," you get "apply to everything except these blocks."

Outputs: block_vector and blocks_range are both lists - the actual sweep, one entry per configuration generated - and summary is a single string readout, useful for sanity-checking what a sweep is about to run before you burn GPU time on it.

How to install it

Through ComfyUI Manager: search "SP-Nodes," install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/bananasss00/ComfyUI-SP-Nodes

Restart ComfyUI afterward. This node needs a Flux LoRA and a compatible block-weight-aware loader downstream to actually do anything with its output - it only produces the configuration.

Common issues

The most common trip-up is the block index list: match the tooltip's format exactly (1,5,6,8, no stray spaces) and stay within Flux's real block counts - 19 double blocks, 38 single blocks - since an out-of-range index is a straightforward way to get a config that silently does nothing useful. And if you were expecting one clean, fixed vector rather than a batch of sweep configurations, you want this node's sibling instead - FluxInspireLbw_BlockVectorPreset - which produces exactly one block_vector rather than a whole set of them.