zhenzhen-pixelcut-video-background-removal-fal

Video background removal is the hard end of a commodity job - per-frame segmentation across motion, hair, and transparency, which is exactly the kind of thing you'd rather not babysit on your own GPU. Comfly_pixelcut_video_background_removal_fal ("zhenzhen-pixelcut-video-background-removal-fal") offloads it to Pixelcut's cloud service: you feed it a video, it comes back with the subject cut out, and you even pick the replacement background - transparent, black, white, green, blue, magenta, or a custom RGB.

It's a FAL-class node in the T8mars/Comfyui-zhenzhen pack, the "Comfly" API-wrapper family that calls the author's paid reseller service. Key goes into api_key directly; FAL billing is pre-deduct and settles after. Submit → poll → download, with the result coming back as a VIDEO tensor.

The inputs that matter

• video_url - required, but ignored when you connect a VIDEO input. So the normal flow is: generate or load a video upstream, wire it in.
• video - the VIDEO input that takes precedence.
• background - what to replace the background with. transparent is the default and the usual answer (composite the cutout in ComfyUI yourself); pick a solid color if you want a ready-made keyable or clean fill.
• custom_r / custom_g / custom_b - the RGB values when background is custom.
• output_format - the useful part. auto (default) or a real codec menu: webm_vp9, mp4_h264, mp4_h265, mov_proresks, mov_h265, mkv_h264, mkv_h265, mkv_vp9, or gif. If you're compositing later, pick something with an alpha-friendly container; if it's going straight to social, h264 is fine.
• video_way - upload (default) or video_url, mirroring how you hand over the file.
• poll_interval / max_poll_attempts - default 6s × 600 = a one-hour timeout ceiling.
• skip_error - fail soft for batch runs.

Outputs

• video - the result as a VIDEO tensor, ready for a save or a composite.
• video_url - the hosted result as a string.
• response - the raw payload.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen
# restart ComfyUI

Or ComfyUI Manager → "Comfyui-zhenzhen", then a key from the author's site into api_key.

The honest part

This is the node that saves you from the worst part of video work: per-frame masking with wobbly hair edges. Cloud processing means your source video - possibly long - leaves your machine and crosses a reseller's server, so this isn't the tool for anything under embargo. And video jobs are the expensive end of any API, so the per-call cost is real; check what a minute of footage runs before you batch an entire edit. For a short clip that needs a clean cutout today, it's hard to beat. For a whole project's worth of footage, your local BiRefNet-class tools are still the volume answer.