The Biological Computing Co. launches mischief 1, a video model it says is shaped by living neurons

The Biological Computing Co. (TBC), a startup based in San Francisco, released an image-to-video model called mischief 1 on October 30, timing the launch to Mischief Night, the eve of Halloween. The company bills it as the first publicly available "neuron-derived" AI video model, a label that rests on its own description of how the model was built.

What is on offer

Mischief 1 is not trained from scratch. It is built on Wan 2.2, an open-weight video model, to which TBC adds an optimization layer that it says increases the parameter count by less than 0.1%. The model is offered through the fal and OpenRouter marketplaces. According to Tao Media, it produces MP4 clips at 16 frames per second, in 480p or 720p landscape format, lasting three or five seconds.

The price is $0.005 per second at 480p, so a five-second clip costs 2.5 cents, and $0.01 per second at 720p. TBC says this makes it the cheapest image-to-video model on both platforms. The company also claims generation is five times faster than the base Wan 2.2 at matching quality, with inference costs about 80% lower. Tao Media cites a TBC test averaging 142 seconds per clip for Wan 2.2 versus 12 seconds for mischief 1.

Where the neurons come in

TBC says it grows living neurons in its lab, records how they process information and turns the principles it identifies into proprietary software. The neurons never leave the lab: customer requests run on ordinary AI hardware. The company argues this approach differs from common efficiency techniques such as quantization, distillation or caching, but it has not published a technical description that would let outsiders check that claim.

Every performance figure comes from TBC. No independent benchmark accompanies the launch, Tao Media's coverage presents the numbers as the company's own, and the only outside voice in the release is Justin Hackney, co-founder of Wonder Studios, a customer that tested the model in production. Whether the biological research is what produces the speed-up, rather than more conventional engineering, cannot be verified from the material made public so far.

Context

The launch follows a collaboration with Amazon Web Services announced roughly five weeks earlier. Tao Media reports plans to run the model on AWS Trainium chips, deploy it through Amazon SageMaker AI and list it on AWS Marketplace, a listing described as coming soon. TBC says video is its first commercial use and that it is working on language and physical AI applications.