HOTCUT · Удалить фон

The quickest cutout in the pack, and the one with the most honest trade-offs. HotcutRemoveBackground (it shows up in the menu as HOTCUT · Удалить фон, because the whole pack is in Russian) takes any IMAGE, sends it to the cloud, and returns the subject on a transparent background for about 10 flames. Two inputs, one output, no model to download, no VRAM.

The mechanism is the same shape as the pack's other cloud nodes: upload your image to POST /api/v1/uploads, submit a remove-background task, poll, download, decode. There's genuinely nothing else to it - no quality slider, no "keep the person only" toggle, no resolution picker. You get the service's one best cutout, every time.

Inputs: config and image. That's it. Output: image (IMAGE).

The honest take, though: background removal is the most commoditized job in this whole ecosystem. Local options are free, fast and genuinely good - BiRefNet has shipped in ComfyUI core since May 2026, and rembg and InSPyReNet have been one node away for years. If you have a GPU and a cutout-heavy pipeline, local is the right call, and at any real scale the flames add up: 10 each, and a 1000-image batch is actual money in rubles.

So why would you reach for this node? Because "free and fast" still assumes you've installed the model, picked a checkpoint, and have VRAM to spare. This node wins when you want a cutout now with zero setup, when you're on a weak or no GPU, or when you're already paying for the rest of the pack and one image at a time is fine. It's also consistent by design: one fixed model on the server means no "which weight variant did you use" variance across your workflow.

One real gotcha, and it's the source-vs-README kind. The README promises a transparent PNG, but look at what the node actually does with the response: it decodes the result into a plain RGB IMAGE tensor, so the alpha channel is dropped on the way back into the graph. Don't expect a true RGBA layer you can stack with clean edges - the subject comes back cut out, but you'll be compositing it on a flat background rather than carrying real transparency downstream. If that matters, the local BiRefNet path gives you proper alpha and costs you nothing per image.

Install is the pack standard: ComfyUI Manager → search "HOTCUT" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Smyshnikof/comfyui-hotcut.git
cd comfyui-hotcut
pip install -r requirements.txt

And the same single-frame note as the Edit node: feed it a batch and only the first frame goes up. For the occasional one-off cutout, ten flames and zero setup is a fair price. For anything batch or production, this is the node to skip - BiRefNet is right there in core.