Pika Labs launches Pika Experiments with Generative UI, a voice-driven interface

Pika Labs launches Pika Experiments with Generative UI, an open-source voice interface using OpenAI's Realtime API and Pika MCP to build dynamic HTML.

Pika Labs shares its experiments with Pika Experiments. A series of raw prototypes released open source, with a first project named Generative UI. The principle: a voice-controlled interface where an AI agent listens to the conversation in real-time and dynamically composes the most suitable display for each response, without using predefined templates.

Specifically, the system generates a complete HTML layout with each interaction, which can take the form of a product sheet, an image grid, a list of logos, a video player, or a calendar. The demonstration follows a consistent thread: starting from a voice request about a product concept, the agent produces a sheet, several renderings, logo proposals, then an advertising video with music, before creating an event in Google Calendar.

Technically, the voice relies on OpenAI's Realtime API, while creative functions are handled by Pika MCP, an in-house tool layer comprising about sixty functions, from image and video generation to lip-syncing and music. With each response, the model outputs a JSON interface description that the server converts into HTML and CSS in real-time. A Pika account connection is required, with Google Workspace integration remaining optional.

The project is shared on GitHub as a local prototype, to be run on one's own machine. Pika Labs itself presents it as experimental and unfinished.