BMAB Lora Loader

If you're already deep in BMAB's world of binds and its own internal pipeline, this is the LoRA loader built for it. It doesn't patch a MODEL/CLIP pair directly the way core ComfyUI's LoraLoader does - instead it produces (and optionally consumes) a BMAB lora type, which is BMAB's own way of describing "here's a LoRA and its strengths" as a value that gets chained through the graph and applied later, wherever BMAB's pipeline actually does the patching.

How it works

The chaining is the interesting part. Wire a previous BMAB LoRA Loader node's lora output into this one's optional lora input, and you're stacking - each node in the chain adds one more LoRA at its own strength_model/strength_clip, and the whole accumulated stack travels forward as a single BMAB lora value. This is the same idea as chaining multiple LoraLoader nodes in a vanilla graph, just expressed through BMAB's own type instead of repeatedly re-patching MODEL/CLIP. Where that final chain gets consumed - which BMAB node actually applies the accumulated LoRAs to your model - isn't one of the nodes covered in this batch, so if you're building a graph from scratch, look at BMAB's example workflows (linked from the README) to see which downstream node expects a BMAB lora input.

The inputs and outputs that matter

• lora_name (enum, required) - populated from your ComfyUI models/loras/ folder, same as core's loader. If the dropdown is empty, you don't have any LoRA files where ComfyUI expects them.
• strength_model / strength_clip (default 1, range -100 to 100) - the usual two dials: how strongly the LoRA affects the diffusion model versus the text encoder. Negative values invert the LoRA's effect, which is occasionally useful for suppressing a trait rather than adding it, but it's a niche move - start positive.
• lora (optional, BMAB lora) - feed in a previous BMAB LoRA Loader's output here to stack onto it. Leave empty if this is the first (or only) LoRA in your chain.
• Output: lora (BMAB lora) - pass to the next BMAB LoRA Loader to keep stacking, or on to whatever BMAB node consumes the final stack.

How to install it

ComfyUI Manager: search comfyui_bmab. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/portu-sim/comfyui_bmab.git
cd comfyui_bmab
pip install -r requirements.txt

No extra model downloads for the node itself - obviously you still need whatever .safetensors LoRA files you want to load, placed in models/loras/ like any other ComfyUI LoRA.

Common issues & troubleshooting

lora_name dropdown is empty. Your LoRA files aren't in the folder ComfyUI is actually scanning. Check extra_model_paths.yaml if you're using a custom models directory, and restart ComfyUI after adding new files - the dropdown is populated at startup, not live.

Stacking multiple LoRAs and the result looks muddy or over-cooked. This is a general LoRA-stacking problem, not specific to BMAB: strengths don't average, they compound. Two LoRAs at 1.0 each can easily overshoot what either does alone. Drop each one's strength_model toward 0.6–0.8 when stacking rather than assuming 1.0 is always safe.

The LoRA doesn't seem to be doing anything at all. Confirm the chain actually reaches wherever BMAB applies it - an unconsumed BMAB lora output sitting on a dead-end wire has no effect on your generation no matter what strength you set. Trace the wire all the way to the node that consumes it.