TrimBG Advanced

Background removal is the most commoditized operation in the whole ecosystem - BiRefNet's the default, it's even in ComfyUI core now, and it gets you a clean subject mask in under a second. But a removed background leaves you with a full-size canvas and the subject floating in a sea of nothing. TrimBG Advanced (from comfyui-missed-tool) is the cleanup step everyone forgets: it crops the canvas down to your subject's bounding box, so a 1024×1024 cutout becomes just the object, tight.

It's the "Advanced" sibling of the pack's plain TrimBG. The simple version takes an image and auto-crops; this one takes a mask too, plus padding, an alpha threshold, and an optional background fill. It also handles a whole batch, which is the difference between doing one image and doing a hundred.

How it works

Give it an image and a mask, and it composites the subject onto a transparent background, finds the alpha's bounding box, and crops to it. The four knobs are where the control lives:

• mask - your cutout mask (from BiRefNet, a segmentation node, or this pack's Load Image Alpha). If you leave it disconnected, it falls back to using the image's luminance as alpha - handy, but don't expect magic.
• padding - expand (positive) or shrink (negative, down to −1000) the crop around the subject. Negative padding is how you shave the mask's sloppy halo off the edges.
• alpha_thresold - spelled exactly that way in the code, a classic author typo. Any pixel with alpha below this value (0–255) is zeroed out before the bounding box is computed. Raise it to cut wispy, half-transparent edges out of the crop.
• fill_rbg - a hex background color (e.g. #ffffff) to flood into the remaining transparent pixels, so you get a solid-background image instead of RGBA.

The gotchas that will actually bite

• fill_rbg needs the #. The code only applies the fill when the string is longer than 6 characters - #ffffff works, but ffffff silently does nothing. Easy to miss; type the hash.
• It crops, it doesn't change the content. Semi-transparent areas inside the subject (glass, hair, smoke) that survive the threshold are kept - the fill only touches fully transparent pixels. If you need fractional-alpha handling, that's a matting model's job, not a crop node's.
• Batch pairing. When images come in as a batch, the node pairs each with a mask in order. If your masks come from a node that emits them one at a time instead of as a batch, you'll get mismatched crops - feed both as batches.

Where it fits in a real workflow

The pack author's own recommendation: use the Inspire Pack's Load Image List From Dir to read a folder as a list, BiRefNet (ComfyUI_BiRefNet_ll is the one they name) to build masks, then this to trim. That's a batch pipeline that goes folder → subject cutouts → compositing, and TrimBG Advanced is the node that makes the output presentable instead of a full-canvas float.

Install

Standard for the pack - ComfyUI Manager, search comfyui-missed-tool, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ifmylove2011/comfyui-missed-tool.git

Restart and it's under missed-tool as "TrimBG Advanced". No extra Python dependencies (the pack runs on torch/PIL, already in ComfyUI) - just remember you'll need a segmentation model or mask source upstream to actually use the mask input.