Load Video LoRA (URL/Path) & Apply

Despite living in the "Video" section of this pack, this node's required inputs are just model and clip - the exact same pair the stock ComfyUI LoRA loader patches. That's the tell: this one is built for native ComfyUI Wan support (the official example workflows, no third-party wrapper involved), not for Kijai's WanVideoWrapper pipeline. If you're running Wan the way ComfyUI ships it out of the box, this is the LoRA stacker that matches your setup; if you're on WanVideoWrapper instead, you want this pack's other Wan LoRA node, LoadVideoLoraFromUrlOrPathSelect, which outputs the wrapper-specific WANVIDLORA type instead of a patched model/clip pair.

What it does is functionally the same job as chaining several stock LoRA loader nodes together, except each slot fetches its LoRA from a URL instead of requiring the file already sit in your loras folder, and it handles up to ten of them in one node instead of ten separate ones cluttering your graph.

Inputs that matter

• model / clip (required) - the pair to patch, straight from your checkpoint loader.
• toggle (default on) - disables the whole node's effect for quick before/after comparisons.
• num_loras (1–10) - how many slots are active.
• lora_N_url per slot - the LoRA's URL or local path.
• lora_N_strength_model / lora_N_strength_clip per slot - separate strength dials for the model and the CLIP side, same split you'd get from the stock LoraLoader node, just multiplied across up to ten LoRAs in one place.

Outputs are MODEL and CLIP, patched and ready to feed straight into your sampler and text encoder the same way they would coming out of any other LoRA node.

Installing it

Search ComfyUI-load-lora-from-url in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/bollerdominik/ComfyUI-load-lora-from-url

You'll need a Wan checkpoint already loaded through native ComfyUI (the official Wan example workflows are the reference point for how that pipeline is wired) feeding model and clip into this node.

Where people get burned

• Confusing this with the wrapper-pipeline version. This is the single most likely mistake with this node specifically: if your workflow uses Kijai's WanVideoWrapper nodes anywhere upstream, this node's MODEL/CLIP output won't line up with what the rest of that pipeline expects. Check which Wan loading path the rest of your graph is actually on before picking which of the two Wan LoRA nodes in this pack to use.
• Stacking Wan speed LoRAs (lightx2v, CausVid and similar) at full strength. These are genuinely useful for cutting generation time down from tens of minutes to a few, but the well-documented tradeoff is that they flatten motion, lighting and skin detail if left at strength 1.0 across the board. The common workaround is applying the speed LoRA at reduced strength, and only where it's actually needed for speed, rather than as a default on every LoRA slot.
• A Wan LoRA trained for the wrong model size or variant. Wan ships in multiple parameter sizes (1.3B, 5B, 14B) and architectures (dense vs. MoE across 2.1/2.2), and a LoRA trained against one doesn't necessarily transfer to another the way you'd hope - check what the LoRA's own page says it was trained against before assuming it'll just work.