Load HunyuanLora From HF

Same job as this pack's other Hunyuan LoRA loaders - it produces a HYVIDLORA object for Kijai's ComfyUI-HunyuanVideoWrapper to consume - just sourced from a public HuggingFace repository instead of a CivitAI ID. If you don't have HunyuanVideoWrapper installed, this node's output has nowhere to go; install that first.

What it's for

Research-adjacent Hunyuan LoRAs and adapters often land on HuggingFace straight from whoever trained them, ahead of (or instead of) a CivitAI re-upload. If that's where the file you want lives, this node fetches it directly by repo and filename at run time rather than making you download and place it manually - the same value proposition as every other loader in this pack: useful the moment your ComfyUI instance doesn't already have the file sitting on disk, which is the normal situation for API workers and freshly-spun cloud executors.

The inputs and outputs that matter

• repo (STRING, required) - the HuggingFace repository name, user/repo form, per the node's own tooltip.
• file_name (STRING, required) - the exact filename inside that repo, matching path and extension.
• strength (FLOAT, default 1, range -10 to 10, step 0.0001, required) - "LORA strength, set to 0.0 to unmerge the LORA."
• prev_lora (HYVIDLORA, optional) - the standard HunyuanVideoWrapper stacking pattern: chain LoRAs by feeding one's lora output into the next one's prev_lora.
• blocks (SELECTEDBLOCKS, optional) - restrict the LoRA to specific transformer blocks rather than the whole model, an advanced option most workflows skip.
• Output: lora (HYVIDLORA) - feeds the next node's prev_lora, or the wrapper's model loader directly.

One gap worth flagging: unlike the plain LoadLoraFromHF node elsewhere in this pack, there's no token-carrying variant of this one. If the Hunyuan LoRA you want sits in a private or gated HuggingFace repo, this node has no field for that - you're limited to public repos here.

How to install it

• ComfyUI Manager - search "comfyui-model-dynamic-loader," install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/jax-explorer/comfyui-model-dynamic-loader, then restart.

You also need ComfyUI-HunyuanVideoWrapper installed separately for the output to connect to anything. Neither pack bundles the (much larger) Hunyuan base-model weights themselves.

Common issues & troubleshooting

404 or repo-not-found. Verify repo is exactly user/repo and file_name matches the file's actual path inside the repo, including extension - a .safetensors typo or wrong subfolder will fail the fetch the same way a wrong URL would.

It's a gated or private repo. There's no auth here - this node can only reach public repos. If you control the repo and it's gated, you'll need to fetch it another way (or ask whoever maintains this pack about adding a token variant, since the plain-HF loader elsewhere in the pack already has one).

Nothing to plug the output into. Install ComfyUI-HunyuanVideoWrapper - HYVIDLORA belongs to that pack, not stock ComfyUI.

LoRA has no visible effect. Confirm the file is actually trained for Hunyuan and for the version you're running; a LoRA for an unrelated architecture won't error on load, it just won't do anything. Trigger words in the prompt are also worth checking against whatever documentation shipped with the LoRA.

Slow first run. That's the download over network, not a hang - with no local cache described in the README, every fresh call against a cold executor re-fetches the file.