San Francisco startup says lab-grown brain cells helped it make an AI video tool faster
The Biological Computing Co. grows living neurons in a lab, studies how they handle information and turns what it finds into software. Its first product makes short AI videos for half a cent a second.
The Biological Computing Co., a San Francisco startup, released an AI tool Oct. 30 that turns a still picture into a short video, according to the company and the AI marketplace OpenRouter. The company says lab-grown brain cells helped design it.
The tool, called mischief 1, is built on Wan 2.2, an open-weight video model, meaning anyone can download and use it for free. The company says its version makes video five times faster and costs 80% less to run than the original. Those claims have not been independently tested.
How brain cells handle information
Neurons are the brain's messengers. They use electrical and chemical signals to pass information around the brain and the body, according to the National Institute of Neurological Disorders and Stroke, part of the National Institutes of Health. One neuron passes a message to the next by releasing chemicals called neurotransmitters across a tiny gap called a synapse.
Brains do this with very little energy. A National Institute of Standards and Technology researcher wrote in 2023 that the human brain can do the equivalent of a billion billion math operations a second on about 20 watts of power. One of the world's most powerful supercomputers, Frontier at Oak Ridge National Laboratory, needed about a million times as much power to match that pace, he wrote.
How the company uses them
The company grows living neurons on a chip covered with tiny electrodes, small metal contacts that can both send and pick up electrical signals, according to its website. Researchers turn pieces of video into patterns of electrical pulses, send them into the cells and record how the activity spreads through the network over time.
They then look for patterns in the timing and in how the neurons affect one another, and turn promising patterns into computer code. Each change is tested against the original AI model to see whether it runs faster, costs less or works better. The company says its work uses dishes of cells, not whole brains, and that there is no possibility of consciousness.
No living cells are in the product. "The neurons stay in our lab," the company says on its website. What customers get is a small add-on that fits around each of the model's main building blocks. It adds less than 0.1% to the model's parameters, the millions or billions of adjustable numbers a model uses to make its output, and it runs on ordinary AI chips.
The company has described earlier tests on its blog. In one, images of handwritten numbers were passed through living neurons before reaching a simple AI program, which then recognized the numbers more accurately than it did on its own. In another, an add-on based on the neurons helped an AI that generates Minecraft game worlds keep long videos from falling apart, the company said.
Price and backers
OpenRouter lists mischief 1 at half a cent per second of video at standard resolution and 1 cent at high definition, in clips of three or five seconds. At the lower price, a five-second clip costs 2.5 cents. It is also offered on fal, another AI service.
The company was founded by Alex Ksendzovsky, its chief executive, and Jon Pomeraniec, its chief operating officer, both neurosurgeons and neuroscientists. It raised $25 million in February in a round led by the venture firm Primary and opened a lab in San Francisco's Mission Bay neighborhood, it announced then. In September, it announced it is working with Amazon Web Services, Amazon's cloud computing business, to offer the model to businesses. On Oct. 15, it said TIME magazine named its technology one of the Best Inventions of 2026.
What could come next
The company says video is only its first use. It is applying the same process to newer video models, with early results showing bigger gains, and is working on language AI, the kind behind chatbots, and "physical AI," such as robots, it said in its Oct. 30 announcement.
Its long-term goal is what it calls "real-time biological computing," in which living neurons and silicon chips work together on AI tasks. "Video is our first commercial application," Ksendzovsky said in the announcement. The goal, he said, is "building more capable models without simply adding more compute."
Source: OpenRouter, "mischief 1" model page, released Oct. 30, 2026; The Biological Computing Co., launch announcement, Oct. 30, 2026; company technology page; About page; company blog; AWS announcement; TIME honor announcement; seed funding announcement, Feb. 12, 2026; NINDS, "Brain Basics: The Life and Death of a Neuron"; NIST Taking Measure blog, March 15, 2023.
- OpenRouter: TBC mischief 1 model page
- The Biological Computing Co.: launch announcement for mischief 1 (Oct. 30, 2026)
- The Biological Computing Co.: Technology page
- The Biological Computing Co.: About page
- The Biological Computing Co. blog: Biological Neural Dynamics for Computer Vision
- The Biological Computing Co. blog: Bridging Biology and Silicon: The Neural Dynamics Adapter
- The Biological Computing Co.: AWS collaboration announcement (Sept. 23, 2026)
- The Biological Computing Co.: TIME Best Inventions announcement (Oct. 15, 2026)
- The Biological Computing Co. seed round release via PR Newswire (Feb. 12, 2026)
- National Institute of Neurological Disorders and Stroke: Brain Basics: The Life and Death of a Neuron
- NIST Taking Measure blog, Advait Madhavan: Brain-Inspired Computing Can Help Us Create Faster, More Energy-Efficient Devices (March 15, 2023)
- The Biological Computing Co.