☁️BizyAir Wan_LoraLoader

This node doesn't read your local models/loras folder - nothing in this pack does, since the actual generation happens on BizyAir's servers, not yours. Instead it pulls from LoRAs you've uploaded to your BizyAir account. The pack's own changelog documents this directly: BizyAir added support for running custom LoRA models back in September 2024, followed shortly after by the ability to share them, and this node is the ComfyUI-side control for that feature.

It only does one thing on its own - output config that another node consumes. Pair it with BizyAir Wan_ImageToVideoPipeline, whose optional lora_config input is exactly where this node's output goes.

Why this matters for Wan specifically

Character consistency is the biggest reason people reach for LoRAs on Wan at all - training a face or a style on 20–30 stills with something like musubi-tuner and applying it across generations is the standard workflow in the wider Wan ecosystem. A useful thing to know if you're bringing an existing LoRA over: LoRAs trained on Wan 2.1 carry forward cleanly to 2.2, so you're not locked to whichever version you originally trained against.

Inputs and outputs that matter

• lora_name - a dropdown that populates from LoRAs registered to your BizyAir account. If you haven't uploaded or shared anything yet, it'll show a placeholder rather than a real option - there's nothing local to fall back on here.
• model_version_id (default empty string) - this looks like it's referencing the specific version identifier of your uploaded LoRA on BizyAir's model-hosting service, separate from the display name in lora_name. If your account has multiple versions of the same LoRA uploaded, this is presumably how you pin a specific one - though the pack doesn't document the field in detail, so treat that as an educated read of the intent rather than confirmed behavior.
• lora_weight (optional, default 1, range 0–1) - the standard strength dial, per the node's own tooltip translating to "LoRA weight strength." Lower it if the LoRA is overpowering the base generation; most people land somewhere in the 0.6–1.0 range depending on how strongly the LoRA was trained.

Output is a single lora_config, which only makes sense wired into BizyAir Wan_ImageToVideoPipeline's matching input - it doesn't do anything on its own.

How to install it

• ComfyUI Manager - search "BizyAir", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git, restart.
• Comfy CLI - comfy node install bizyair.

Set your API key before doing anything with this pack - the in-app login prompt or bizyair.siliconflow.cn. You'll also need to actually upload or otherwise register a LoRA to your account through BizyAir's own model-hosting flow before it shows up here; installing the node pack alone doesn't give you anything to select.

Common issues & troubleshooting

lora_name shows nothing to pick. This means your BizyAir account doesn't have a LoRA registered yet, not that the node is broken. You need to upload or share one through BizyAir's model-hosting feature first - this loader only lists what's already attached to your account.

A LoRA you trained locally doesn't show up. It has to be uploaded to BizyAir specifically; a LoRA sitting in your local ComfyUI loras/ folder isn't visible to this node at all, since generation runs cloud-side.

Applying the LoRA doesn't visibly change output. Check lora_weight first - if it's set low, or the pipeline node's lora_config input isn't actually connected, you'll get a generation that looks like the LoRA was never applied. Confirm the wire is there before assuming the LoRA itself is weak.

The pack turned up without you deliberately installing it. A ComfyUI Manager bug used to mis-resolve missing nodes onto BizyAir, and that's the source of most "is this malware" concerns you'll find about it. It's a real, legitimate commercial pack - remove it through the Manager if you don't want it, keep it if you're using nodes like this one.