Image Remove Bg

You want a subject cut out of its background, cleanly, without leaving ComfyUI. That's the whole job here. easy imageRemBg bundles four different background-removal models behind one node and a dropdown, so you can flip between them on the same image and keep whichever handled your hair, fur, or fuzzy edges best. That "try them all on your own image" habit is exactly what the community landed on, because nobody has actually won background removal - it's the most commoditized operation in the ecosystem and still the least settled.

What it does and how it works

Under the hood each mode is a segmentation model that predicts a mask separating foreground from background. Feed it an image, get back the cut-out image plus the raw mask. The four rem_mode choices matter more than anything else on the node:

• RMBG-1.4 (default) and RMBG-2.0 - BriaAI's models. Fast, solid on simple subjects. But read the licensing note below before you ship anything.
• Inspyrenet - the quiet favorite. The highest-scoring background-removal thread of 2026 calls it "absolute magic," and it genuinely wins on some material that RMBG smears. It's MIT-licensed, so it's the safe commercial pick.
• BEN2 - newer, launched with vendor benchmarks beating the incumbents, then more or less stopped being talked about. Worth a click on a hard image; not worth standardizing on.

Notably, this node doesn't offer BiRefNet, which is the model ComfyUI now ships in core and the usual default recommendation elsewhere. If you specifically need BiRefNet HR for a 2000px product shot, you'll reach past this node. For everyday cutouts, the four here are plenty.

The inputs and outputs that matter

images is your input. rem_mode is the model. After that:

• refine_foreground - cleans up the edge halo (color bleed from the old background). Leave it on for anything with hair.
• add_background - none keeps transparency; white or black flattens onto a solid color, handy when the next node can't handle an alpha channel.
• torchscript_jit - a speed toggle for Inspyrenet. One of the few quantified head-to-heads in the wild ran Inspyrenet "with torchscript_jit on," so if you're batching, turn it on.

image_output (Preview/Save/Hide) and save_prefix are the usual output-node plumbing. The two outputs are image (the cut-out, wire it into a compositor or save node) and mask, which you can route into an inpaint or paste-back step if you'd rather composite yourself.

Installing it

Easiest path is ComfyUI Manager: search ComfyUI Easy Use, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

Then install the requirements (double-click install.bat on Windows, or pip install -r requirements.txt) and restart. The background-removal weights download on first use, so the first run of any mode stalls while it fetches the model - that's normal, not a hang.

Where people get burned

The big one is licensing, and it's the default that bites you. RMBG-1.4 and RMBG-2.0 are BriaAI weights under non-commercial research terms. They work great, they're the default, and the community reads "free" and stops. If you're doing anything you plan to sell - a product-shot pipeline, a paid service - switch rem_mode to Inspyrenet, which is MIT and clears you. The model quality difference is small; the license difference is not.

Second: no background remover survives everything. Flyaway hair, semi-transparent fabric, and thin structures are where all four struggle, and none of these is a matting model, so a genuinely see-through veil or smoke will come out as a hard-edged blob. Run your three hardest images through each mode before you commit to one - that ten-minute check is the actual state of the art here, not any single model's benchmark.