ByteDance has released Bernini V2, an open-source omnimodal video editor that can edit any existing video using natural language. The release follows the V1 version that ByteDance introduced a few weeks ago.

With Bernini V2, users can take an existing video and edit it with plain-language commands. Examples from the demo include adding two characters to a scene and changing its tone to warmer, removing a specific character or object, changing the camera perspective, or changing the background. Users can also add reference images of things they want to insert into the video, such as placing a helicopter into a scene.

A very large model. Bernini V2 is a massive model at around 180GB just for the weights, not counting the VAE and text encoder. The team notes this makes running it locally difficult, and the first V1 version has not drawn much attention from the open-source community.

Philippine angle. Bernini V2 is an open-source developer release, so there is no retail price or availability in the Philippines. It joins other open video-editing tools that Filipino developers can run locally if they have the hardware.

Source

Bernini V2 specs:

  • Type: open-source omnimodal video editor
  • Editing: natural-language commands on existing video
  • Features: add/remove characters and objects, change camera perspective and background, insert reference images
  • Model size: around 180GB (plus VAE and text encoder)
  • License: Apache 2.0
  • Status: open-source release, no consumer product announced