☁️BizyAir Shared Load ControlNet Model

This is BizyAir's ControlNet loader, and the interesting part isn't the model list - it's the word "Shared" in the name and the share_id field that comes with it. That's not a standard ComfyUI ControlNet loader input; it looks like a BizyAir-specific mechanism for reusing an already-loaded model across multiple places in your graph (or across runs) by key, instead of re-uploading or re-loading the same multi-hundred-megabyte ControlNet checkpoint every single time. That's a sensible thing to want when your compute lives on someone else's server and every load is a network round trip, not a local disk read.

What's actually in the model list

Eight checkpoints, and the spread tells you something honest about which architectures BizyAir has actually built ControlNet support for:

• kolors/Kolors-ControlNet-Canny.safetensors
• kolors/Kolors-ControlNet-Depth.safetensors
• sdxl/diffusion_pytorch_model_promax.safetensors
• sd15/control_v11f1e_sd15_tile.pth
• instantid/diffusion_pytorch_model.safetensors
• sd3.5_large_controlnet_blur.safetensors
• sd3.5_large_controlnet_canny.safetensors
• sd3.5_large_controlnet_depth.safetensors

That's Kolors, SDXL, SD1.5, the InstantID ControlNet, and SD3.5 - but notably nothing for Flux, despite the pack's changelog advertising Flux ControlNet support elsewhere (BFL shipped official Flux Depth and Canny models back in late 2024). If your workflow needs a Flux ControlNet specifically, double-check whether this particular loader has one before assuming it does - the list above is what's actually exposed as of this writing, and it's SDXL-and-earlier-heavy.

Inputs and outputs

• control_net_name - pick from the eight checkpoints above. Match this to your base model family and your intended conditioning type (Canny for edges, Depth for spatial layout, Tile for detail-preserving upscale workflows, InstantID's for face structure).
• share_id - a string, default "share_id". This is the caching key discussed above. If you're using two ControlNet passes in the same graph that reference the same checkpoint, giving them the same share_id likely lets BizyAir reuse the already-loaded model server-side instead of loading it twice - worth testing if you're chasing execution speed in a graph with repeated ControlNet loads.

Output: CONTROL_NET, typed BIZYAIR_CONTROL_NET - wires into BizyAir's ControlNet Apply node, not a vanilla ComfyUI ControlNet apply node.

How to install it

Ships with the full pack.

ComfyUI Manager: search "BizyAir," Install, restart.

git clone:

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

API key required before this (or any BizyAir node) will actually load a model.

Common issues

Model/architecture mismatch. This is the single most common ControlNet mistake in general, not a BizyAir quirk: a Kolors ControlNet will not condition an SDXL generation, and an SD1.5 tile model won't work against SD3.5. Match the checkpoint's prefix in the dropdown to your base model exactly.

No Flux option here. As covered above - if you came looking for Flux ControlNet conditioning specifically, this loader's list doesn't currently have it, even though the pack overall lists Flux support in its changelog for other features (Flux Fill, Flux Kontext, and so on). Don't assume every architecture BizyAir supports elsewhere also has a ControlNet checkpoint in this particular loader.

share_id collisions across unrelated graphs. Since it defaults to the same static string for everyone, if BizyAir's caching is scoped more broadly than per-run, it's worth giving it a distinctive value when you actually need isolation - though for a typical single-user workflow the default is unlikely to cause visible problems.

API key / credits. As with every BizyAir cloud node, a missing or exhausted API key surfaces as a load failure here rather than a descriptive local error.