Latest in AI
TBC Mischief 1: Video API Pricing, Resolution and What Neuron-Derived Means
TBC’s new public video model starts at half a cent per second, with a biological research claim that does not require biological hardware.
The Biological Computing Co. announced public access to Mischief 1 on October 30 through fal and OpenRouter. The image-to-video model starts at $0.005 per second for 480p output, offering a low-cost way to generate short clips from a starting image and a text prompt.
TBC describes Mischief 1 as neuron-derived. That needs explanation before it becomes the headline people remember: customers are using software on conventional AI hardware, not sending video requests to living neurons.
Mischief 1 pricing and availability
The fal model page lists different prices for the two available resolutions. The half-cent figure applies to 480p, not every generation.
| Output | Price per requested second | Three-second clip | Five-second clip |
|---|---|---|---|
| 480p | $0.005 | $0.015 | $0.025 |
| 720p | $0.01 | $0.03 | $0.05 |
At these rates, 100 five-second clips cost $2.50 at 480p or $5 at 720p, before separate application costs. The difference matters if your product defaults to the higher resolution.
OpenRouter also lists the model, with an October 30 release date and the same resolution-dependent pricing. Its page says EU in-region routing is not currently available. That is a specific routing restriction, not a claim that European users cannot access the model at all.
What neuron-derived means here
According to TBC’s announcement, its researchers measure living neurons in the lab and turn computational insights from those experiments into software optimizations. Mischief 1 is built on Wan 2.2, with the company reporting an addition of less than 0.1% to the base model’s parameter count.
TBC claims five-times-faster generation while matching the base model’s quality. These are company claims, not independently verified conclusions established by the announcement. The biological origin of an optimization does not by itself establish quality, reliability or superiority over other optimization methods.
The product question remains ordinary: what useful video do you get for the time and money spent? The research story is interesting, but it does not replace a workload-specific comparison.
Clip limits and API behavior
The fal API schema supports three- or five-second clips at 16 frames per second, in landscape 480p or 720p. It requires an image URL and a prompt, and returns an MP4 video.
The default prompt-processing mode enhances the supplied prompt before rendering. For repeatable comparisons, the documentation says to select original as well as fix the seed, because the rewriting step does not take a seed.
That is a small setting with a large evaluation consequence. Otherwise, two runs intended to compare generation behavior may also be comparing different rewritten instructions.
The most plausible early use is exploring variants before committing to a final shot. Lower per-attempt cost makes that easier, but teams should measure the cost per usable clip, including rejected generations and review. That is the number that determines whether inexpensive video generation becomes an inexpensive workflow.
I’m Paras. I work at the intersection of AI, engineering and distribution.
I help technical founders turn complex launches into clear stories that explain both the technology and its practical limits.