Load Lora From HuggingFace

Functionally this is ComfyUI's standard LoRA loader - the one that patches your model and CLIP with low-rank weight deltas - except the file comes from a public HuggingFace repository instead of your local models/loras folder. You give it a repo and a filename, it fetches, applies, done.

When to reach for it

HuggingFace has become the place a lot of LoRAs and adapters get published alongside their base models, especially in the video and newer-architecture space where CivitAI's catalog hasn't caught up yet - think Wan or Hunyuan LoRAs shared straight from a research repo rather than re-uploaded to a fine-tune marketplace. If the LoRA you're after lives on HuggingFace and the repo is public, this node saves you the manual download-and-place step, which matters most when you're not running ComfyUI on a machine with a persistent, pre-populated model folder - an API worker or a fresh cloud instance being the obvious cases.

If the repo you need is private or gated (license click-through, access request, invite-only), this node can't reach it - that's what its sibling, LoadLoraFromHFWithToken, exists for.

The inputs and outputs that matter

• model (MODEL) and clip (CLIP) - required, the pair the LoRA patches.
• repo (STRING) - the HuggingFace repository name in user/repo form, per the node's own tooltip.
• file_name (STRING) - the exact filename inside that repo, per the tooltip. This has to match the file as it actually sits in the repo's file tree, including its extension.
• strength_model and strength_clip (FLOAT, default 1, range -100 to 100, step 0.01) - the usual two independent dials for how hard the LoRA pushes the diffusion model versus the text encoder. There's no single correct value across the board: the old SDXL-era convention of keeping this around 0.5–0.8 doesn't hold everywhere, and some current character LoRAs are published with a recommended weight of 1.0 or above because turning it down costs likeness. Whatever the LoRA's own page or card recommends beats a default guess.
• Outputs: MODEL and CLIP, wired onward exactly like any LoRA loader's outputs.

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. No dependencies or bundled downloads listed in the README - the LoRA file itself is the only thing that gets pulled, and only when a workflow actually calls this node.

Common issues & troubleshooting

404 or "repository not found." Double-check the repo string is exactly user/repo with no leading slash or trailing path, and that file_name matches a real file in that repo's tree - a wrong subfolder or a .bin where the repo actually stores .safetensors will fail the same way.

It's a gated or private repo. This node has no auth field at all, so a repo that needs a login simply won't resolve here - use LoadLoraFromHFWithToken instead, which accepts a HuggingFace access token for exactly that case.

LoRA loads without erroring but does nothing visible. LoRAs are bound to the architecture they were trained on; loading one built for a different base model family than the checkpoint you're running is silent, not an error, so check the repo's model card for what it's meant to pair with. Also check for a trigger word - a lot of LoRAs need one in the prompt to actually kick in.

Slow on first run. That's the download, not a stall - HuggingFace file sizes for LoRAs typically land in the tens to low hundreds of megabytes, and with no local cache to fall back on, every fresh run against a cold executor pays that cost again.