🌀 BV Vector of Length-n

Block-weight tuning is the power move of LoRA work: instead of one global strength, you give every block of the model its own weight, so a LoRA can be strong where it helps and zeroed where it hurts. Every one of those schemes starts with a plain list of floats - one per block - and BV Vector of Length-n is how you create that list out of thin air.

If you're on FLUX, here's the shape you're aiming at: 58 slots - the input layer, then the 19 MMDiT double-stream blocks, then the 38 single-stream DiT blocks. The BV FLUX-specific nodes hardcode exactly that layout by name, which is a strong hint this whole vector family was built for the block-weight workflow on FLUX. vector_length = 58, default_value = 1.0, and you have a neutral all-ones vector ready to edit.

Inputs

• vector_length (default 1, min 1) - how many entries. For a FLUX block-weight vector that's 58; for an SDXL one it's a different number, and for anything custom it's whatever your consumer node expects.
• default_value (default 0, stepped 0.01) - the value every entry starts at. 1.0 means "LoRA fully applied everywhere"; 0.0 means "off everywhere," which is a fine starting point for selectively turning blocks on.

Output

• VECTOR (LIST) - a list of floats of the requested length, all set to default_value.

What it's for

This is the constructor. On its own it does nothing interesting; its job is to be the input to the edit nodes - BV Vector Edit to change one slot, BV Vector Edit Range for a span, the FLUX dropdown variants if you'd rather pick blocks by name. The typical session: build a 58-long vector, punch in the blocks you care about, then hand the result to a block-weight node or convert it to a string with BV Vector to String.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/BlackVortexAI/ComfyUI-BVortexNodes.git

or ComfyUI Manager → "BV Nodes", then restart. No dependencies, no models.

Gotcha

There's no length checking against the consumer - if you build a 50-long vector and feed a node that wants 58, that's a mismatch you'll discover at runtime, not something this node prevents. Build the length to match your target model's block count and you'll be fine.