← All experiments
Generative media

R24

An autonomous 24-hour tech news channel, with a newsroom and on-air media generated seconds before broadcast.

R24 · Broadcast and newsroom

The experiment

R24 explores what it takes to keep a generative media experience running continuously. Instead of preparing a fixed sequence of clips, a newsroom selects stories and creates material close to airtime.

The format combines an AI anchor, real press images, and generated on-location correspondents.

How the newsroom works

  1. Gather. RSS feeds and grounded search supply the newsroom’s inputs.
  2. Decide. A Gemini newsroom determines coverage and allocates short or longer segments, including 20-second and 90-second formats.
  3. Generate. AI presenters and correspondents deliver the selected stories alongside press imagery.
  4. Broadcast. Reactor FastH3 generates material seconds before air.

Why continuous generation matters

A channel is an ongoing system. Its editorial decisions and generated output need to connect over time. R24 brings those pieces together in a concrete format: a tech news broadcast that runs itself.

That makes it a useful companion to our world-model work, with a focus on orchestration and real-time generative media.

What we’re exploring

  • Coordinating retrieval, story selection, and media generation.
  • Adapting a story to different segment lengths.
  • Maintaining a consistent broadcast format across changing stories.
  • Producing material close to the moment it is shown.

Project notes

R24 is a studio artifact. Its source repository includes setup instructions and the broadcast workflow.

Project links

For live demos with interactive media

Try live demos.