(WIP) Merged LoRA → WanVideo

Start with the honest framing: this node's own display name in ComfyUI is literally "(WIP) Merged LoRA → WanVideo" - the author is telling you upfront it isn't finished. Worth knowing before you build a graph around it.

What it's trying to do

If you're running Wan through kijai's WanVideoWrapper (a WANVIDEOMODEL, not a standard MODEL), the wrapper can expose fewer keys than the core model does. That's a problem for LoRA Optimizer, because its merge output is computed against the full set of weight keys - if the wrapper hides some of them, patches meant for those keys have nowhere obvious to land. This node is meant to bridge that gap: it takes the merged patches from LoRA Optimizer and pushes them into the wrapper model via its own set_lora_params mechanism, so patches the wrapper would otherwise not expose still reach the sampler.

The reality, from the author directly

This isn't a guess - the pack's own author addressed it directly in the Reddit thread where the pack was first shared, when someone asked about using it on Wan LoRAs: "it's a node for the wrapper but it's not working correctly, i'll probably remove it if I can't fix it. the normal node work with code wan 2.2 lora." In other words: if you're on standard Wan 2.2 with the regular MODEL/CLIP path (not the WanVideoWrapper), you don't need this node at all - the normal LoRA Stack → LoRA Optimizer path already works there. Reach for this bridge specifically if you're committed to the WanVideoWrapper ecosystem, and go in expecting rough edges rather than a polished feature.

The inputs and outputs that matter

• wan_model (required) - a WANVIDEOMODEL, from a WanVideoWrapper model loader, not a standard checkpoint loader.
• lora_data (optional) - the LORA_DATA output from LoRA Optimizer (or AutoTuner, Merge Selector, Extract from Model - anything that produces one).

Output: model - a WANVIDEOMODEL, wire it onward to your WanVideoSampler in place of the input.

How to install it

Same as the rest of the pack - ComfyUI Manager, search "LoRA Optimizer", install, restart. Manually:

cd ComfyUI/custom_nodes/
git clone https://github.com/ethanfel/ComfyUI-LoRA-Optimizer.git

Restart - it lands under loaders. You'll also need ComfyUI-WanVideoWrapper installed separately, since WANVIDEOMODEL is that pack's type, not this one's.

Common issues & troubleshooting

You're on plain Wan and don't actually need this node. If your graph uses the standard checkpoint loader and produces a regular MODEL, skip this entirely - LoRA Stack → LoRA Optimizer already works on core Wan 2.2 LoRAs without any bridging.

Patches don't seem to reach the sampler, or behavior is inconsistent. This matches what the author himself flagged - the WanVideoWrapper bridge is known not to work correctly as of this writing. Don't assume you've misconfigured something; this is a documented rough spot in the node itself, not a setup mistake.

Considering this for a production workflow. Given the node is explicitly marked WIP and the author has floated removing it outright if it can't be fixed, treat any workflow depending on it as provisional. If the WanVideoWrapper path matters to you, keep an eye on the pack's changelog before building something you depend on around this specific node.