JLC 2-LoRA Loader - Block Weight [Legacy]

Straight answer up front: JLC 2-LoRA Loader - Block Weight is a compatibility node, and the author says so. It exists so older saved workflows that reference the old fixed two-slot class still open and run after the pack refactored its LoRA family. For anything new, you should use the pack's dynamic loaders instead - this one is explicitly slated for deprecation in a future release. If you're building a fresh workflow and someone dropped this node into a recipe, swap it for the current-generation equivalent and move on.

So when do you use it? When you open an old workflow and it's already there, and you just want it to keep working without rebuilding the graph. That's the whole job.

What it does

It's a fixed two-LoRA loader that patches both the MODEL and the CLIP/text encoder, with a shared block-weight vector applied to the model side. "Block weight" is the LoRA idea where you don't apply a LoRA uniformly - you scale how strongly it patches different blocks of the network, so a LoRA can be all-in on the early input blocks and near-zero on the output blocks, for example. The block_vector is a numeric CSV string with one value per block, defaulting to 1,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1. Zeros suppress matching patches, non-zeros scale them, and the helper recognizes block families from both UNet-era (input_blocks, middle_block, output_blocks) and Flux-era (double_blocks, single_blocks) architectures.

The inputs

• model and clip - both inputs and both outputs, since it patches each side.
• block_vector - the shared CSV model block-weight vector described above.
• lora_01 / lora_02 - the two LoRAs, from your models/loras folder.
• strength_01 / strength_02 - legacy combined strength, used for both MODEL and CLIP.
• strength_model_XX / strength_clip_XX - per-side overrides. Set either one and it takes precedence over the combined strength for that side.

Outputs: model and clip, straight into your sampler.

Installing it

It ships inside the jlc-comfyui-nodes pack, so:

cd ComfyUI/custom_nodes
git clone https://github.com/Damkohler/jlc-comfyui-nodes.git

or search jlc-comfyui-nodes in ComfyUI Manager, then restart. No extra dependencies, no downloads.

The honest upgrade path

The reason this node is legacy: the pack's new dynamic LoRA family handles up to ten slots, treats the slot count as authoritative with hidden rows kept safely in the workflow JSON, and splits block weighting into a shared variant (one vector for all slots) and a per-slot variant (each LoRA gets its own vector). The closest modern replacements for this node's job are the "Multi-Model / Block Weight" loaders - grab the shared-block-weight MODEL+CLIP one if you want the same one-vector-for-two-LoRAs behavior, and the per-slot one if you were manually juggling two profiles. When you do migrate, note that the new family's block weighting applies to the MODEL only, same as here; CLIP patching stays ordinary per-LoRA strength. Old workflows keep working in the meantime - but treat that as a courtesy, not a feature to build on.