𝙆 BiRefNet Loader

If you landed here confused because you already read about KayTool's BiRefNet Loader and this looks identical - it is, and that's the point of this article. KayTool actually registers two loader nodes with the exact same display name, the exact same inputs, and the exact same output: one under the class name BiRefNet_Loader, and this one under Kay_BiRefNet_Loader. "BiRefNet Loader" is a common enough name that it's very likely to collide with another pack's own node of the same name - several other packs in the ecosystem ship their own BiRefNet integrations - and Python-level class-name collisions between custom node packs are a real, known landmine in ComfyUI (whichever pack loads second can silently fail to register, or overwrite the earlier one). Shipping a second copy under a prefixed class name is a reasonable way to dodge exactly that.

Functionally, treat this node as identical to KayTool's other BiRefNet Loader - same model choices, same behavior, same paired Remove BG node downstream. This entry exists so you have somewhere to land if this is the specific class name your workflow (or a JSON you imported) references.

How it works

Loads a BiRefNet model and outputs it as a REMOVE_BG-typed value for KayTool's Remove BG node to consume. model chooses the weight:

• BiRefNet - general-purpose default.
• BiRefNet_HR - higher native resolution, better for large input images where the base model would otherwise downscale internally and lose edge detail.
• BiRefNet-portrait - specialized for people.

providers chooses the hardware backend - auto, cpu, cuda, or mps - same as the sibling node.

The inputs and output that matter

• model (choice of BiRefNet / BiRefNet_HR / BiRefNet-portrait).
• providers (choice of auto / cpu / cuda / mps).
• REMOVE_BG (output) - feeds into Remove BG.

Installing it

Same as the rest of the pack: ComfyUI Manager → search KayTool → install → restart, or cd ComfyUI/custom_nodes && git clone https://github.com/kk8bit/KayTool and restart. Both loader nodes ship in the same install - there's nothing extra to do to get this specific one. Model weights download from HuggingFace on first actual use, not at install time.

Common issues

If a workflow JSON you've imported references Kay_BiRefNet_Loader specifically and the node shows up red/missing after installing KayTool through Manager, double-check you're on a recent enough version of the pack - this class name looks like a later addition alongside the original BiRefNet_Loader, so an outdated install might only have one of the two. Beyond that, the troubleshooting is identical to the other loader: a stalled first run is almost always the model download (check your connection and disk space), and if providers: auto isn't landing on your GPU, set cuda explicitly. If you're not sure which of the two loader nodes a given workflow actually needs, either one works the same way - wire in the model and provider choice you want and connect its REMOVE_BG output to Remove BG exactly as you would with the other.