The Biological Computing Co. announced Mischief 1 on October 30, a video-generation software product based on Wan 2.2. The company says measurements of living neurons informed modifications to the model. Customer requests run on conventional AI hardware; the cells remain in the laboratory.

Product pages on fal and OpenRouter confirm public service listings. They do not independently verify the manufacturer’s claimed fivefold acceleration.

What users can actually request

OpenRouter describes image-and-text input and video output, with 480p and 720p options, three- or five-second clips and a 16:9 aspect ratio. At the time of checking, its page listed prices starting at $0.005 per second for 480p and $0.01 for 720p.

A five-second clip therefore corresponds to $0.025 or $0.05 at those listed rates. This is a calculation based on generated duration, not a promise about the final cost of an editing project. Retries and the number of variants affect total expenditure.

The same page says the model is not currently available through EU in-region routing. That is different from saying EU users cannot access it at all. For an organisation that requires regional data processing, the distinction matters.

Biological origin, digital product

In an interview with The AI Innovator, the co-founders describe observing cellular activity, extracting computational patterns and translating them into software. The report discusses a component called the Neural Dynamics Adapter that modifies processing within an existing model.

That is the company’s explanation as conveyed in a journalistic interview, rather than an independently reproduced experiment. It does not mean the service contains a human brain. Proposed future work involving direct biological computation is separate from the current commercial software.

The interview and the current product listing provide different pricing information. For the practical figures above, we use the checked provider page without assuming a promotional window or one price across every configuration.

Turning the speed claim into a useful comparison

Our editorial assessment is that a video creator’s most relevant comparison is time and cost to an acceptable shot. If a faster variant requires more retries to preserve an object or its motion, acceleration of an individual generation will not necessarily translate into the same improvement across a complete project.

A meaningful test would hold the starting image, duration and resolution constant, then compare several outputs. The person judging them should decide in advance what matters: shape consistency, motion or adherence to the instruction. This is a proposed evaluation method, not a benchmark SIExpo has performed.

It would also be misleading to compare a short low-resolution request with a longer high-resolution one and attribute the entire difference to model design. Settings and the acceptance criterion belong beside any reported speed figure.

Public availability makes the announcement a concrete service development. Stronger claims about efficiency and a biologically derived advantage still require evaluation under specified conditions. Keeping those questions separate preserves the interesting development without treating its unusual origin as a quality guarantee.

Cover: conceptual AI illustration, not a photograph of an actual cell culture, equipment or measured result.

Editorial check: 2026-11-01T09:38:54+02:00 (Europe/Sofia).