Hunyuan Multi-Lora Loader (Wrapper)

This is the sibling node in the Hunyuan-Multi-Lora-Loader pack, and it exists for one reason: Kijai's ComfyUI-HunyuanVideoWrapper doesn't want you chaining LoRA nodes at all. Its HunyuanVideo Model Loader node accepts LoRA information in a format called HYVIDLORA, and this wrapper node produces exactly that. Same four slots as its sister node, completely different job.

The key difference to internalize up front: this node has no model input and no model output. It does not apply a single weight. All it does is build a list of LoRA descriptors and hand them to a loader that does the applying. If that sounds like it should be the fiddly variant, you're right - but it's the one you want if you're living inside Kijai's wrapper workflow, which most HunyuanVideo users are.

How it works

Under the hood it's embarrassingly simple, and that's the point. For each of the four slots it reads the LoRA name, strength, and block type, and - as long as the slot isn't set to "None" and strength isn't 0 - appends a small dict to a list:

{
    "path": <full path to the .safetensors>,
    "strength": <your strength>,
    "name": <filename without extension>,
    "blocks_type": <all / single_blocks / double_blocks>,
    "blocks": None
}

That list comes out of the lora output, typed HYVIDLORA, and you wire it into the lora input of Kijai's HunyuanVideo Model Loader. The node's own description says it plainly: it outputs "LoRA information in HYVIDLORA format for compatibility with HunyuanVideo Model Loader." No model is touched, so there's nothing to re-run or cache incorrectly - if you change a slot, the output list changes and the loader picks it up.

The inputs that matter

Same shape as the main loader, so if you've used one you know the other:

• lora_01..04 - dropdown of ComfyUI/models/loras, "None" to disable.
• strength_01..04 - default 1.0, -10 to 10. Zero disables the slot.
• blocks_type_01..04 - all, single_blocks, double_blocks. These get passed through to the loader.

One honest caveat about blocks_type here: the wrapper just carries the value in the dict - it does no key filtering of its own, unlike the main loader node. Whether flipping it from all to double_blocks actually changes your output depends on how Kijai's loader interprets the field, which lives outside this repo. If it doesn't seem to do anything on the wrapper path, that's the loader's business, not this node's. The pack's author recommends double_blocks in the main node; on the wrapper, treat it as a preference you pass along and test.

Which node should you use?

That's the real question, and the answer is about your loader, not about LoRAs:

• Native ComfyUI HunyuanVideo workflow (built-in HunyuanVideo loader nodes) → HunyuanMultiLoraLoader, the model-in/model-out one.
• Kijai's wrapper (HunyuanVideo Model Loader) → this wrapper node, feeding its lora input.

Don't cross the streams. The wrapper's HYVIDLORA output is not a MODEL - it will not plug into the main loader node, and the main loader's model output won't satisfy the wrapper's expectations. They solve the same "four LoRAs in one place" problem in the two different worlds HunyuanVideo runs in.

Installing

Identical to its sibling - both nodes ship in the same pack, so one install gets you both:

# ComfyUI Manager: search "Hunyuan-Multi-Lora-Loader", install, restart
# or:
cd ComfyUI/custom_nodes
git clone https://github.com/asdrabael/Hunyuan-Multi-Lora-Loader

No extra dependencies, no pip installs, no model downloads - it's a single Python file using ComfyUI's built-in folder helpers. It only becomes useful once you have a HunyuanVideo setup already running, so install order is: get Kijai's wrapper working first, then add this pack to tidy up the LoRA part. And since this is a one-author, essentially zero-community pack, treat it as a convenience layer: if the wrapper's own LoRA handling ever improves, the wrapper node is the part you can drop without losing anything else.