Nodes/SP-Nodes/FluxInspireLbw_BlockVectorPreset
ComfyUI Node

FluxInspireLbw_BlockVectorPreset

One fixed block-weight vector for a Flux LoRA

By bananasss00·Created 2 years ago·Updated 10 days ago· 20
FluxInspireLbw_BlockVectorPreset
    • block_vector
    enable_double_blocks
    enable_single_blocks
    reverse_modefalse

    The simpler sibling of FluxInspireLbw_Batch. Where that node sweeps through a whole set of block-weight configurations for testing, this one commits to exactly one, fixed. LoRA Block Weight - applying a LoRA at different strengths per network block instead of one flat strength - is an old A1111-era idea most associated with ltdrdata's Inspire Pack; this pack ports the concept to Flux's own transformer structure, which splits into 19 double-stream blocks (joint image/text processing) and 38 single-stream blocks (the combined stream after that), unlike the old UNet's IN/MID/OUT split the original technique was built around.

    How it works

    Once a FluxInspireLbw_Batch sweep (or your own experimentation) tells you which blocks actually matter for a given LoRA, you hardcode that answer here as a stable, reusable preset instead of regenerating a whole sweep every time you load that LoRA.

    • enable_double_blocks - a comma-separated list of double-block indices to enable, in the format the tooltip spells out: 1,5,6,8.
    • enable_single_blocks - same idea, for single-block indices.
    • reverse_mode - flips the selection to "everything except these blocks" instead of "only these blocks."

    Output: a single block_vector string, not a list - this feeds into a compatible block-aware LoRA loader elsewhere in the graph the same way the batch version's individual entries do, just without the sweep.

    Worth noticing what this node doesn't have compared to FluxInspireLbw_Batch: no block_strength or forced_block_strength field. This node only decides which blocks get enabled, not how strongly - the strength presumably falls to whatever default the downstream block-aware loader applies. If you need explicit per-strength control alongside your block selection, that's a reason to reach for FluxInspireLbw_Batch instead, even if you only care about its first generated entry.

    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.

    Common issues

    Both enable_double_blocks and enable_single_blocks default to an empty string, which means it's easy to leave them blank and then be confused about why the LoRA seems to apply everywhere (or nowhere, depending on how the downstream loader treats an empty vector) - double-check you've actually typed indices into at least one of the two fields before assuming something else is broken. Beyond that, the same formatting rule as the batch node applies: match the 1,5,6,8 tooltip format exactly, and stay within Flux's real block counts (19 double, 38 single) - an index outside that range won't do anything useful.

    One more thing worth flagging for anyone new to block-weight tooling generally: block indices aren't consistently numbered across every implementation of this idea, so a block list you find in a guide written against a different tool (Inspire Pack's SDXL vectors, for instance) won't transfer here - Flux's double/single split is a different numbering scheme entirely, and the only reliable way to know which indices matter for a given LoRA is to sweep them yourself with FluxInspireLbw_Batch first.

    CategorySP-Nodes

    Inputs (3)

    NameTypeDefaultDescription
    enable_double_blocksSTRINGexample: 1,5,6,8
    enable_single_blocksSTRINGexample: 1,5,6,8
    reverse_modeBOOLEANfalse

    Outputs (1)

    NameTypeDescription
    block_vectorSTRING