After four years of researching living brain cells, TBC has incorporated a neuron algorithm into Alibaba's Wan 2.2, claiming video generation is 5 times faster.

Beating AI News Flash: TBC, a US company founded by two neurosurgeons, has released the video model Mischief 1.

The new model is adapted from Alibaba's open-source Wan 2.2, adding less than 0.1% parameters. TBC claims that with comparable image quality, video generation is at least 5 times faster and inference costs are reduced by about 80%.

This technology comes from the team's four years of research on living neurons. Researchers cultured brain cells on a chip with 4,096 electrodes, converted visual information into electrical stimulation, and recorded how neural activity propagates through space and decays over time. Based on these patterns, they designed lightweight software modules and embedded them into the diffusion Transformer structure of the video model to help the AI maintain consistency between frames without retraining the entire model from scratch.

Video models can be sped up by reducing denoising steps, but if the number of steps is reduced too much, images tend to become blurry and distorted, and it becomes harder to maintain consistency between frames. TBC previously conducted experiments using an AI model capable of generating real-time Minecraft visuals based on player actions. After adding bio-inspired modules, the model could reduce denoising steps while still maintaining image stability. Combined with caching and sampling optimization, the throughput of the entire system increased by 4.4 times.

Now, TBC is pushing this research direction toward commercialization. Mischief 1 has been integrated into fal and OpenRouter and currently supports generating 3-second or 5-second videos from images. Generating a 5-second 480p video costs only 2.5 cents, while 720p costs 5 cents.