Load WanVideoLora From ComfyOnline

Two things have to be true for this node to be useful to you. First, you need Kijai's ComfyUI-WanVideoWrapper installed - the output type is WANVIDLORA, which that pack defines and its model loader consumes; stock ComfyUI has no idea what to do with it. Second, you need an account on ComfyOnline (comfyonline.app), the paid ComfyUI cloud service this whole pack is built around, since the comfyonline_model_id field expects an ID from that platform's own model library rather than a public source like CivitAI or HuggingFace.

If you're not already on ComfyOnline, this specific node has nowhere to fetch from - its siblings in this same pack, LoadWanVideoLoraFromCivitAI and LoadWanVideoLoraFromHF, cover the sources anyone can browse without an account.

What it's for, if you are on that platform

WanVideoWrapper's own LoRA-select node points at a local file already downloaded into your LoRA folder. This node swaps that for an ID lookup against ComfyOnline, folding the fetch into the run itself - the same pattern as every other loader in this pack, useful specifically when you're not running on a machine with the file already on disk.

The inputs and outputs that matter

• comfyonline_model_id (STRING, required, default empty) - "The ID of the model to download from ComfyOnline." Where exactly to find this ID isn't documented in the pack's README; it's presumably whatever identifier the platform's own model library assigns.
• strength (FLOAT, default 1, range -10 to 10, step 0.0001, required) - "LORA strength, set to 0.0 to unmerge the LORA." If what you're loading is a speed/distillation LoRA (LightX2V, CausVid and similar), the community-tuned range for Wan 2.2 is 0.4–0.6 rather than full strength, applied on the low-noise pass specifically - pushing speed LoRAs harder than that is a known way to flatten motion and lighting quality, not just make generation faster.
• prev_lora (WANVIDLORA, optional) - chains multiple LoRAs: feed one node's lora output into the next node's prev_lora before the last one reaches WanVideoWrapper's model loader.
• blocks (SELECTEDBLOCKS, optional) - restricts the LoRA to specific transformer blocks instead of the whole model. An advanced, rarely-needed knob fed from a block-selection node elsewhere in WanVideoWrapper.
• Output: lora (WANVIDLORA) - into the next node's prev_lora, or straight into the wrapper's model loader if it's the only one in the chain.

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-WanVideoWrapper installed separately for the WANVIDLORA output to connect to anything. Neither pack bundles Wan's own base-model weights, which are a much larger separate download handled through your normal Wan setup.

Common issues & troubleshooting

The ID field doesn't resolve to anything. Expected if you don't have a ComfyOnline account - there's no documented public fallback here. For a public-source alternative, use LoadWanVideoLoraFromCivitAI or LoadWanVideoLoraFromHF instead.

Nowhere to plug the lora output. Install ComfyUI-WanVideoWrapper first - WANVIDLORA belongs to that pack.

Motion or detail quality tanks after adding a LoRA. If it's a speed LoRA, you're most likely running it too hot - try 0.4–0.6 on the low-noise pass instead of full strength across the board, a widely reported Wan 2.2 pattern that has nothing to do with which node fetched the file.

LoRA has no visible effect at all. Confirm it was actually trained for Wan - a LoRA meant for a different architecture loads without error and simply does nothing.