Bria muscles video background removal with V-RMBG 3.0

Bria launches V-RMBG 3.0, an autoregressive video background removal model offering four API endpoints, 40 ms real-time streaming, and full IP indemnity.

Bria, a specialist in visual AI infrastructure trained exclusively on licensed data, is upgrading its video background removal suite to V-RMBG 3.0. The model is based on a new autoregressive architecture that processes video with temporal awareness to eliminate the flickering and unstable edges inherent in frame-by-frame approaches, recurring weaknesses with hair, motion blur, and semi-transparent objects.

The suite covers four API endpoints: classic background removal with 256-level alpha matte, virtual green screen available in three chroma key colors, background replacement with an image or video, and a real-time streaming mode announced at around 40 ms inference per frame, designed for live commerce, AI avatars, or streaming. Bria claims a success rate exceeding 63% against its direct competitors and speeds up to nine times faster, figures derived from its own benchmarks. The upgrade applies automatically to existing customers via the current endpoints, without migration or minimum commitment.

A free sandbox allows testing the entire suite without registration, with deployment options including cloud, BYOC, or on-premise. Every output is covered by full IP indemnification.