☁️BizyAir Shared Lora Loader

Normal LoRA loaders assume the file lives on your machine. This one assumes it lives on someone else's BizyAir account, and you've been given a code to borrow it. BizyAir has had LoRA-sharing baked in since September 2024 - the platform lets people host a custom LoRA and hand out access without ever exporting the raw file - and this node is the loading half of that: point it at a share ID instead of a filename, and it pulls that person's model straight from BizyAir's cloud.

It's genuinely useful for exactly one situation: someone gave you a share code for their custom LoRA (a style, a character, a concept) and you want to run it without asking them to upload a multi-hundred-MB safetensors file, or without needing anywhere to store it yourself. Outside that situation - if you already have the LoRA file locally - this isn't the node you want; you'd use a regular loader instead.

How it works

Give it a share_id, and it resolves that ID against BizyAir's hosted models, then applies the resulting LoRA the same way any LoRA loader does: blending its weights into a model and a CLIP at whatever strength you set.

Inputs and outputs

• share_id - a string, the code identifying the shared LoRA. This is the one field the whole node revolves around; get it wrong and nothing downstream matters.
• lora_name - an enum that starts with zero choices until a valid share_id populates it. If this dropdown stays empty, that's your first sign the share ID isn't resolving.
• model and clip - both typed BIZYAIR_MODEL and BIZYAIR_CLIP, meaning they have to come from other BizyAir nodes upstream (a BizyAir checkpoint or diffusion-model loader, and a BizyAir CLIP loader). You can't wire in ComfyUI's stock MODEL/CLIP outputs here - the types simply don't match, so keep this part of your graph as an all-BizyAir chain.
• strength_model and strength_clip - floats from -100 to 100, both defaulting to 1. These work like any LoRA strength control: 1.0 is full effect, 0 is off, negative values invert the LoRA's effect (occasionally useful for pushing away from a style rather than toward it), and values above 1 overdrive it.

Outputs are MODEL and CLIP, both still typed BIZYAIR_MODEL/BIZYAIR_CLIP - pass them on to the next BizyAir node in your chain (a sampler-facing node, or another LoRA loader if you're stacking).

Installing BizyAir

Through ComfyUI Manager: search "BizyAir," install. Or by hand:

cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git

then restart ComfyUI. Comfy-CLI works with comfy node install bizyair. As with every node in this pack, you need to set an API key first - the README's "click to login" link walks you through registering. Loading a shared LoRA is inherently a cloud operation (you're pulling a file that isn't yours and was never meant to land on your disk), so there's no local fallback if the key's missing.

Common issues

An empty lora_name dropdown means the share_id didn't resolve - double-check you copied the whole code and that whoever shared it with you actually still has it shared (sharing can presumably be revoked on their end, same as any hosted-access model).

Because both model and clip need to come from BizyAir-typed sources, a common beginner trap is trying to feed this a checkpoint loaded with a standard ComfyUI loader - that connection will simply refuse to make in the graph editor. If you're building a workflow around a shared LoRA, plan for the whole chain (checkpoint load, CLIP, sampling) to stay inside BizyAir's node family rather than mixing in your local loaders partway through.