Load Video LoRA (URL/Path) Wan

The output type here, WANVIDLORA, is your first clue about what this node is actually for: it's not a general LoRA loader, it's built specifically to feed Kijai's ComfyUI-WanVideoWrapper, the third-party wrapper that became the standard way to run Wan video with extras like SageAttention and TeaCache baked in. WANVIDLORA is a type that wrapper defines and consumes - nothing else in stock ComfyUI knows what to do with it. So before this node is useful at all, you need WanVideoWrapper installed and its own LoRA-apply node downstream of this one.

What it buys you over just using WanVideoWrapper's native LoRA loading is the same trick as everywhere else in this pack: instead of downloading each Wan LoRA file and placing it in the right folder before you can even open the workflow, you paste the URL straight into the node.

The fields you'll actually touch

• toggle (default on) - kill switch for the whole node, handy for quick A/B testing with and without your LoRA stack.
• num_loras (1–10) - how many of the ten slots are active.
• lora_N_url and lora_N_strength per slot - paste the LoRA link, set its weight. Unlike the pack's general-purpose LoRA stacker, there's no separate model/clip split here - Wan video LoRAs patch the video model directly through WanVideoWrapper's own mechanism, so one strength value per slot is all there is.
• prev_lora (optional, type WANVIDLORA) - chain multiple of these nodes together the same way the pack's other stackers chain, feeding one node's output into the next node's prev_lora.

Output is a single lora, type WANVIDLORA, which goes into whatever WanVideoWrapper node applies a LoRA list to the model (commonly something like a WanVideo Lora Select node in that pack).

One node in this same repo is easy to confuse with this one: LoadVideoLoraFromUrlOrPath (no "Select" in the name) does a related job but outputs standard MODEL/CLIP instead - that's the version for people running Wan through native ComfyUI support rather than through Kijai's wrapper. Check which pipeline you're actually building before picking one.

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 also need ComfyUI-WanVideoWrapper (Kijai) installed and a Wan checkpoint already loaded through it - this node's output is inert without that pipeline downstream.

Where people get burned

• No WanVideoWrapper installed. The WANVIDLORA output will have nothing willing to accept it, and you'll get a red "missing node" or type-mismatch error rather than a clean failure. Install the wrapper first.
• Speed LoRAs stacked at full strength on both passes. If your Wan LoRA is a distillation/speed LoRA (lightx2v, CausVid, and similar), the community's own hard-earned advice is that stacking these at strength 1.0 across the whole generation flattens motion, lighting and skin detail into something noticeably worse - the standard workaround is to apply the speed LoRA only on the low-noise pass, at reduced strength (roughly 0.4–0.6), and keep the high-noise pass clean. That's a Wan-architecture thing, not specific to this node, but it's exactly the mistake this node makes easy to make since it'll happily stack ten LoRAs at strength 1 with no warning.
• A Wan 2.1 LoRA that "shouldn't" work on Wan 2.2. In practice it usually does - Wan 2.1 LoRAs are broadly forward-compatible with 2.2 - but it's worth knowing that's an empirical community finding rather than an official guarantee, so test before trusting it blindly on something important.