The pitch is odd enough to get attention: an image-to-video model that runs faster because of what researchers learned from living neurons. The Biological Computing Co., a San Francisco startup, released TBC Mischief 1 on Thursday. The speed claim is big. The evidence so far is the company’s word.
Mischief 1 turns still images into short video clips. It is built on Wan 2.2, Alibaba’s open video model, and it is available on the developer platforms fal and OpenRouter. TBC says it runs about five times faster than Wan 2.2 at the same quality. Pricing is $0.005 per second of video, so a five-second clip costs 2.5 cents.
That price is the part developers will care about. Image-to-video is a crowded field, with Runway, Luma, Pika, Kling and Google’s Veo all competing, and cost per clip is a real constraint for anyone building a product on top of these models. Wan 2.2 is already a popular open option because you can host it yourself. A version that is meaningfully cheaper and quicker, available through an API you already use, would be easy to try.
But the key caveat is right there in the announcement. TBC has not published independent benchmarks. "Same quality" is a subjective claim in video, where motion consistency, artifacts and prompt adherence vary a lot between samples. Five times faster could mean a lot of things depending on resolution, clip length and hardware. Until third parties run side-by-side tests against stock Wan 2.2, treat the number as marketing.
The neuron angle is also less sci-fi than it sounds. The living cells stay in TBC’s lab. The company studies how lab-grown neuron networks respond to signals, then turns those findings into a small software add-on that runs on ordinary AI chips. So you are not renting a dish of brain cells. You are getting an optimization technique, with a biological origin story. Whether that origin is doing real work or just making a good narrative is the open question, because plenty of speedups in this space come from distillation, quantization and better sampling, none of which need a biology lab.
Jon Pomeraniec, TBC’s co-founder and chief operating officer, framed the goal broadly: "We’re building a platform that can change the economics of AI at scale." That is a business pitch, not a technical one, and it fits what TBC has been doing commercially. The launch follows a September deal with Amazon Web Services to sell its models to businesses, which gives the company a distribution channel beyond developer marketplaces.
The wider context is a growing interest in biology as a path to cheaper AI. Inference costs are the thing squeezing every video and image model provider, and energy use keeps climbing. Some groups want neurons to do the computing directly, and a rack of living neurons is already running in Singapore. TBC takes the more practical route, using neurons as a source of ideas rather than hardware.
Does it matter? Potentially, but not yet. If independent tests confirm the speedup on a model people already use, the approach could be applied to other open models, and that would be worth watching. If they do not, Mischief 1 is a cheap Wan 2.2 variant with an unusual backstory. At half a cent per second, developers can test it for almost nothing. Run your own comparison before believing either story.
