The Biological Computing Co. plans to sell a text-to-video model enhanced by a tiny software layer designed from measurements of living nerve cells. The startup claims the resulting system can generate video five times faster and at 80 percent lower inference cost than its unnamed open-source base model.
No biological hardware runs in the finished product. Researchers grow cortical neurons on chips with 4,096 electrodes, stimulate the cells and measure how activity spreads and fades. They translate those patterns into a Neural Dynamics Adapter that plugs into a conventional diffusion model and adds less than 0.1 percent to its size.
The company demonstrated the approach with Oasis, a 600-million-parameter Minecraft world model, where accumulating frame errors can make scenes dissolve over time. The adapter is intended to preserve spatial and temporal structure, but the announced commercial video model has not been released for outside evaluation. The company also has not identified its base model.
A partnership with AWS would run the service on Trainium chips, serve it through SageMaker AI and list it in AWS Marketplace. For now, availability is limited to an early-access signup, so the speed, cost and quality claims remain unverified.