🅐 Image Remove Background

Background removal is the most commoditized operation in this whole ecosystem - every tool does it, and most of them are just a UI over the same rembg library. ACE_ImageRemoveBackground is exactly that: a ComfyUI wrapper around rembg (the u2net family and friends), with one extra trick up its sleeve - a bundled local implementation of Bria's RMBG-1.4 (briarmbg) that doesn't rely on the rembg package at all. That's a nice touch, because briarmbg tends to give cleaner subject edges than the classic u2net default.

It's part of 🅐 Ace Nodes (hay86/ComfyUI_AceNodes), a one-author grab-bag. It runs entirely locally - no API, no key, just model downloads on first use.

How it works

You feed in images and pick a model from a dropdown of nine:

• briarmbg - the bundled Bria RMBG-1.4 port, resizes to 1024×1024 internally. Generally the best default in this pack for clean edges on a subject.
• u2net, u2netp (light), u2net_human_seg, u2net_cloth_seg - the classic rembg models. u2net is the historical default; hair and thin structures are its weak spot, as they've been since 2020.
• silueta - same network as u2net, smaller.
• isnet-general-use / isnet-anime - better edge quality; the anime variant is the one for illustration.
• sam - a quantized Segment Anything encoder/decoder pair, the "segment this thing" option.

First use of any model downloads it into ComfyUI/models/rembg. The briarmbg path is separate - it's a real PyTorch model loaded from the core/image/briarmbg.py module.

The gotcha you'll hit immediately

The IMAGE output is not transparent. The briarmbg branch composites the subject onto a solid black background, and the rembg branch returns RGBA where the RGB channels still hold the cutout - slice off the alpha and you have a subject on black too. So don't wire the image output into something expecting a PNG with transparency and expect a clean alpha. The thing you actually want is the MASK output - a proper alpha mask you can feed to compositing, inpainting, or an alpha-aware saver. The pattern is: grab the mask, and if you need transparent output, build it yourself from image + mask.

Installing

ComfyUI Manager → search ComfyUI_AceNodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/hay86/ComfyUI_AceNodes

Then restart. The pack's requirements.txt includes rembg (plus transformers, insightface, soundfile and the rest of the kitchen sink), and rembg on Windows can be grumpy about its onnxruntime dependency - if install chokes, that's the usual suspect.

Honest positioning

The KB's own verdict on this space: u2net-path removal is acceptable but not exceptional on hair and transparent materials, and if edge quality is the goal, BiRefNet-based nodes (ComfyUI ships background removal in core since May 2026) are the current recommendation. This node is a perfectly fine workhorse for product shots and simple subjects, and briarmbg punches above the classic default - but if your subjects have flyaway hair or glass, you'll get visibly better results from a BiRefNet or matting-based node. Know what you're buying: a solid rembg wrapper, not the edge-quality frontier.