☁️BizyAir InstantID Face Analysis

InstantID is the tool that made people seriously ask, back in early 2024, whether character LoRAs were about to become obsolete. (They didn't - a face isn't a whole character, and LoRAs still own body type, outfit, and everything that isn't the face.) It works from a single photo with no training, and its real trick over earlier face tools like IP-Adapter FaceID is IdentityNet: a ControlNet-shaped module conditioned on facial landmarks that forces the output to keep the reference's actual facial geometry, not just its general vibe. This node is the first step in that pipeline - it doesn't apply anything visible on its own, it just produces the face analysis InstantID's IdentityNet needs downstream.

What it does

One setting: provider, and it only has one real option - CUDA. On the original desktop InstantID node, this choice exists because InsightFace's ONNX runtime can execute on CPU, CUDA, or other backends depending on your hardware. Here, running on BizyAir's cloud GPUs, that choice is essentially inherited from the upstream node rather than a live decision you're making - you're always getting CUDA because that's what BizyAir's infrastructure runs on.

Output is a single BIZYAIR_FACEANALYSIS - a cloud-side handle carrying the ArcFace embedding and facial landmark data extracted from whatever reference image gets fed into this analysis step further down the graph. On its own, wiring this node up and running it produces no visible result; it only matters once it's feeding a companion InstantID apply node that actually patches the model.

Installing it

ComfyUI Manager: search BizyAir, install, restart. Manual:

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

Restart after. No InsightFace models to install locally, which is a real point in this node's favor - InsightFace's ONNX face-analysis weights have historically been one of the more painful installs in local ComfyUI setups (C++ build requirements, platform-specific wheels), and routing it through BizyAir sidesteps all of that. A BizyAir account with an API key is still required before this node executes.

Where people get burned

Wiring this up and expecting a visible effect. This is purely a face-analysis step - it feeds a downstream InstantID node, and nothing changes in your image until that companion node consumes its BIZYAIR_FACEANALYSIS output.

Assuming InstantID has moved past SDXL. It hasn't - the reference implementation has been dormant since mid-2024, there's no Flux port, and if you're on a newer architecture, InstantID (and this node with it) simply isn't the tool for that model. It's still the right call on SDXL, Illustrious, or Pony.

Ignoring the licensing rider. InstantID's code is permissively licensed, but the InsightFace face-recognition weights underneath it are research/non-commercial - that restriction rides along with anything this pipeline produces, regardless of what license the surrounding code carries.