AI Video Background Remover — BiRefNet Precision Cutouts

BIREFNET PRECISION NO GREEN SCREEN 12 OR 24 FPS

An AI video background remover separates a foreground subject from every frame of a clip without relying on a green screen or uniform color. EZ Character uses BiRefNet—a bilateral reference network for high-resolution dichotomous image segmentation—to build detailed alpha mattes and return transparent WebM, animated WebP, transparent GIF, and white-background MP4 outputs.

How does BiRefNet remove a video background without a green screen?

Chroma keying decides what to erase by measuring how close each pixel is to a selected color. That works quickly on a controlled green, blue, or flat-color matte, but it can also erase clothing, reflections, or fine strands that share the keyed hue. BiRefNet instead estimates which image regions belong to the foreground subject and which belong to the background.

BiRefNet is a bilateral reference network designed for high-resolution dichotomous image segmentation: separating a visually complete foreground object from everything behind it. Because the boundary is based on subject structure rather than one background color, it can preserve hair, fur, and soft motion-blurred edges that an aggressive color key would destroy.

What happens to each frame in the BiRefNet video pipeline?

The clip is decoded and sampled into PNG frames at the selected 12 FPS or 24 FPS profile.
Each frame receives one independent foreground-segmentation pass, producing an alpha PNG. This is per-frame image segmentation, not a temporal video model.
The alpha-frame sequence is encoded into VP9-alpha WebM, animated WebP, transparent GIF, and an H.264 MP4 composited over white.

Processing frames independently is honest but important: the model does not use neighboring frames to stabilize the edge. Complex motion can therefore show some frame-to-frame shimmer, especially around fast blur or changing occlusion.

AI background removal vs green screen keying

Choose the method from the source you actually have. AI Remover is for ordinary scenes where no clean matte was recorded. Solid Color Keying is the faster, free choice when the background was deliberately made uniform.

When to use AI Remover or Solid Color Keying
Compare	AI Remover (BiRefNet)	Solid Color Keying
Background needed	Any ordinary scene; no green screen or uniform color required	A green, blue, or other uniform-color background
Hair and fine edges	Semantic alpha mattes can retain hair, fur, and soft motion-blurred boundaries	Depends on clean color contrast; keyed hues and spill can remove or tint fine detail
Max clip length	10 seconds	30 seconds
Cost	Uses credits; final cost follows the verified frames processed at 12 FPS or 24 FPS	Free; no credits used
Best for	People, animals, products, and existing footage shot in real environments	Green-screen capture, blue-screen capture, and animation rendered on a flat matte

Already have a clean green screen or flat matte? Use the free video background remover.

How long can the clip be?

AI Remover accepts MP4, WebM, or MOV uploads under 80 MB and up to 10 seconds. The studio checks duration in the browser before upload. At the limit, Efficient 12 FPS processes up to 120 frames; Smooth 24 FPS processes up to 240. Twice the sampled frames means roughly twice the segmentation work, so the two profiles have different time and credit trade-offs.

Which AI background removal output should I use?

VP9-alpha WebM

Use the 8-bit alpha video for compositing and browsers or editors that support transparent VP9.

Animated WebP

Use the 8-bit alpha animated image for web previews and compact transparent motion.

Transparent GIF

Use it for legacy compatibility. GIF has only 1-bit alpha, so semi-transparent edge pixels become hard cutoffs.

White-background MP4

Use the broadly playable H.264 version where transparency is unsupported; it has no alpha channel.

AI Video Background Remover — FAQ

How does BiRefNet separate a subject from an ordinary background?

Each selected video frame is sent through foreground segmentation. BiRefNet estimates the subject boundary from the image itself and returns a PNG with alpha, so the result does not depend on finding and deleting one chosen background color.

Does the AI track the subject across neighboring frames?

No. This implementation applies BiRefNet independently to each sampled frame; it is not a temporal or video-native segmentation model. Independent mattes keep the pipeline predictable, but difficult motion or changing occlusion can produce some edge shimmer between frames.

Why is AI background removal limited to 10 seconds?

AI Remover performs one foreground-segmentation pass per sampled frame. A 10-second clip can require up to 120 passes at 12 FPS or 240 at 24 FPS, so processing time and cost rise directly with frame count. Solid Color mode accepts clips up to 30 seconds.

Which download format preserves soft transparent edges?

VP9-alpha WebM and animated WebP preserve 8-bit alpha for gradual transparency around soft edges. GIF reduces transparency to 1 bit, making each pixel fully visible or fully transparent. The MP4 has no alpha channel and is composited over white.

What footage gives BiRefNet the clearest cutout?

Use a clip with one clearly visible subject, useful separation from the surroundings, and as little occlusion as practical. BiRefNet can retain hair, fur, and motion-blurred boundaries that overlap the background color, although extreme blur or rapid shape changes can still make edges shimmer.

How are credits calculated for AI video background removal?

The studio estimates the clip from its duration and the selected 12 FPS or 24 FPS profile. Final billing uses the renderer-verified number of processed frames, with a one-second minimum; failed jobs are refunded. The Solid Color method remains free.

Remove a video background with AI

Upload your video

Prepare the cutout

Check the moving edges