Bria Remove Video Background (Transparent)

Bria Transparent Video Background is the "cut the subject out of a moving shot" node. Feed it any video and it removes the background, returning the foreground as a stack of cut-out frames plus an alpha mask - and from there you either composite the subject over something else or write a genuinely transparent video. It's one of ComfyUI's built-in Bria partner nodes, so the removal itself is a cloud API call: video goes up to Bria's service through Comfy's API layer, a transparent-webm comes back, and the node splits it into frames and mask on your machine. Billed per second of video, no local model, no VRAM.

Why not just do it locally? Because the KB's background-removal essay is blunt that the free/local path (BiRefNet, rembg, InSPyReNet) is a still-image game - doing it consistently frame-over-frame on video is where local tools fall apart on hair, motion, and temporal flicker. Bria's whole selling point here is temporal stability: the mask doesn't jitter between frames the way running a still-image remover per frame does.

The inputs

Only two, which is the point:

• video - the clip to cut out. Duration is capped at 60 seconds (validated before billing).
• seed - controls whether the node re-runs; results are nondeterministic regardless of seed. The standard partner-node disclaimer.

The outputs - read these carefully

This node doesn't return a single video; it returns two things:

• images - the cut-out frames (foreground, ready to composite).
• mask - the alpha mask for those frames.

The node description gives you the two legit wiring options: connect both to a compositing node, or feed them to Save WEBM to write an actual transparent-video file. If you just save the images output alone, you'll get opaque frames on a black background and wonder what went wrong - you need the mask to make transparency real.

Gotchas

• Both outputs matter. images without mask is a cutout on black, not a transparent video. This is the #1 confusion with this node.
• 60s cap. Longer clips get rejected before billing - trim first.
• Per-second pricing. A full-length clip is real money, so do the math before you send a long take. Use the mask output locally (it's yours once it's back) to keep compositing costs at zero.
• Foreground quality gates everything. The removal is only as good as the subject separation; busy backgrounds and motion blur produce messier masks, same as any remover.

The workflow that shines: generate a subject, remove the background here, then composite it over a generated scene or a green-screen replacement - all inside one graph, with only the removal step costing credits.