Happy Halloween! Startup promises faster AI powered by human brain cells

The American startup The Biological Computing Co. is presenting Mischief 1 as an AI video model born from research with living brain cells. The company says it runs five times faster than the original model. But behind the striking announcement is an important distinction: the brain cells do not make any videos themselves. The finished product is conventional AI software running on standard computer chips.

With the launch of Mischief 1 on October 30, The Biological Computing Co. (TBC) is releasing its first commercial AI model. CEO Alex Ksendzovsky calls the launch proof that biological computing can work as a commercial technology.

On social media, he draws a direct link to four years of research into living neurons. The combination of brain cells and artificial intelligence sounds like a fundamentally new way to make computers work.

The reality is more nuanced. Mischief 1 is based on Wan 2.2, Alibaba’s existing open video model. TBC has adapted it using software that, the company says, stems from experiments with living neurons.

On its technology page, TBC explains what actually happens. Researchers grow neurons on electronic plates fitted with electrodes. They send electrical signals to the cells, then measure how the cells respond.

The researchers look for patterns in those responses that could help make AI models more efficient.

They translate promising findings into algorithms, then test those algorithms in existing AI models to see whether performance improves, generation speeds up, or less computing power is needed.

The company puts it this way:

“The neurons stay in our lab. What leaves is cheaper, faster, and better AI.”

The brain cells stay in the lab. Customers get only software that runs on conventional AI hardware.

That distinction matters. The marketing around Mischief 1 could give the impression that living brain cells are directly involved in generating videos. They are not.

Instead, biological research serves as a source of ideas and a way to identify potential improvements to existing software.

TBC says its research has produced a more efficient version of Wan 2.2. The company claims Mischief 1 generates videos five times faster without sacrificing image quality.

It also says the cost of running the model, known as inference, is 80 percent lower.

According to TBC, the gains come from a small software add-on whose properties are derived from the observed activity of living neurons. The addition increases the size of the original model by less than 0.1 percent.

In a technical post, the company describes how a biologically inspired software module can improve the stability of generated videos. The researchers tested it in part using an experimental model that generates a Minecraft-like world.

That research is not the same as an independent evaluation of Mischief 1. The published experiments therefore do not automatically prove that the claimed speed boost for Wan 2.2 is entirely due to the biological method.

Nor are there yet broad, independent comparisons confirming that Mischief 1 runs five times faster across a range of conditions while maintaining quality.

The key question, then, is not just whether the model is faster, but how much of that improvement can specifically be traced to research with living neurons.

Regardless of the biological claims, Mischief 1 is a real product. The model is available through the developer platforms fal and OpenRouter.

Users can upload an image and use a text prompt to generate a short video.

At 480p resolution, it costs $0.005 per second of video. A five-second clip therefore costs just 2.5 cents. At 720p, the rate doubles to $0.01 per second.

The model supports three- and five-second videos in widescreen format.

Those low prices could appeal to video creators and developers. When testing different scenes or quickly producing multiple versions, every saving counts.

TBC calls Mischief 1 the cheapest image-to-video model on both platforms. That comparison also comes from the company itself and depends on the current offerings and settings of competing models.

TBC’s approach is fundamentally different from that of Australia’s Cortical Labs.

That company builds systems in which living neurons are actually part of the computer processing information.

In these biological computers, cultured neurons perform some of the calculations themselves. At TBC, researchers use neurons to develop ideas for digital algorithms.

Both approaches use living cells, but the way they do so is fundamentally different.

TBC is also developing applications for language models and physical AI, including systems for robots. Through a partnership with Amazon Web Services, the company plans to make its technology available to business customers.

Mischief 1 shows how research into biological neurons could help improve existing AI software. Independent research will have to establish just how significant—and distinctive—that contribution really is.

For now, Mischief 1 is best understood as a commercial AI video model with an unusual research background—not a computer that makes videos using living brain cells.