Professional Model Logic
Unlike other removers, this one uses a "smart" method of detecting solid backgrounds and can be tuned with basic user settings. Uses a combination of SAM3, BiRefNet, and Pixel Picker, as well as a outline darkener to make the best auto background removal without physically doing the work.

Images and Videos
Supports Images and Videos. Videos are converted into an animated lossless WebP.

Fast
You might think this would be slower than other background removers, but I organized it so that it works just as fast as others, if not faster.

Directions

• [ Download ]: ComfyUI Manager (It allows you to install missing nodes)

• [ Download ]: Pixel Color Picker (It's not included in the ComfyUI Manager Repos)
  (Drop the PixelColorPicker.py into your custom nodes folder)

• [ Download ]: KY-Nodes
  (I know, it adds a bunch of advanced stuff 99% you won't use, but they are the only one I found that let's me have somewhat of a custom ffmpeg command.)
  ( cd ComfyUI/custom_nodes; git clone https://github.com/yorkane/ComfyUI-KYNode/tree/main )

• Read the note on what each setting does.

Questions

Is it perfect? No, but test it against others, I think you will stick with this one if you do. This is sort of a WIP, testing different methods to see what works best, refining the technique, but understand that even manually removing backgrounds don't yield perfect results.

System Requirements? 64GB of RAM for 8 second videos.

Why JXL? Fast decode, excellent compression. Until webp2, best format in class.

Why Animated AVIF? Significantly better than webp.

Why FFMPEG KYNodes? For some reason everyone is stuck in 1970, using 50 year old technology for images and KY-Nodes are the only guy I found that let's us save animate AVIF. Plus it's much faster than the actual Animated WebP Node.

BEN is too aggressive, how to lower it? Expand the "BEN" node and turn down the sensitivity or use a different color background.