Overview
mischief 1 is an image-to-video model from The Biological Computing Company, live on OpenRouter since October 30. The vendor describes it as the world's first neuron-derived video model, a claim rooted in four years of work on electrode-cultured brain cells; how biological signals reach the generation pipeline is stated only in the company's own framing, with no reproducible paper behind it.
What can be verified is how it works and what it costs. Hand it a starting image and a prompt and it returns a 3-second or 5-second clip at 480p or 720p, billed per output second from 0.005 dollars. Among the 27 video models OpenRouter tracks, it ranks first on both cost and speed, with a 5-second clip around 0.025 dollars and end-to-end latency near 13 seconds.
That places it in the ideation stage: generating a dozen motion placeholders while writing a script, or rough animated previews for a storyboard, at a cost that barely registers. A 720p ceiling and clips of 3 to 5 seconds already describe bulk drafts rather than finished footage.
Key Features
- Image to video: Takes one starting image and a prompt and returns a continuous clip, so existing artwork can be pushed into motion directly.
- 3 and 5 second clips: Clip length is fixed at 3 or 5 seconds, enough for a shot test or a looping fragment.
- 480p and 720p: Two resolutions, so drafts can stay at 480p and only selected shots move up.
- Per-second pricing: From 0.005 dollars per second, about 0.025 dollars for a 5-second clip, so trial cost scales by the clip.
- One OpenRouter key: No separate signup, since calls go through an OpenRouter API key alongside text models on the same billing.
- Top of cost and speed ranks: First on both cost and speed among the 27 video models on OpenRouter, with about 13 seconds of end-to-end latency.
Use Cases
- Scriptwriters generate motion placeholders in bulk to settle pacing before production.
- Storyboards get rough animated previews that show a client how a shot moves.
- Ecommerce teams produce atmosphere clips in a dozen variants per product.
- Moodboards and creative pitches use moving images instead of stills.
- Small tools chain it behind a text model on OpenRouter for one-click clips.
- Games and animation test variants by running several prompts in parallel.
Pros
- From 0.005 dollars per second, the cheapest tier among current video models.
- Fast turnaround of about 13 seconds end to end, which suits repeated iteration.
- Runs through OpenRouter with a single key, no separate account.
- Two resolutions separate the cost of drafts from slightly better versions.
- Image-to-video input means one picture is enough to start.
- Inexpensive enough that generating in bulk at the concept stage costs little.
Pricing
Billed per second of output video from 0.005 dollars, so a 5-second 480p clip costs about 0.025 dollars. Calls go through OpenRouter and settle on the platform's normal billing cycle, with no subscription or minimum. Requests are served by a single upstream provider, Ambient, with about 85 percent availability over the last three days and a median 13-second end-to-end latency. The neuron-derived description exists only as a public company statement with no third-party reproduction.
Summary
Most video models compete on quality leaderboards. mischief 1 picks a different axis by pushing the cost of a single clip down to a couple of cents, which turns generating motion into something you do casually. At that price it does not need to win on final quality, only to carry the dozens of attempts that early concept work requires. As for the neuron-derived origin story, treat it as positioning until someone reproduces it.
Version History
- mischief 1 listed: from 0.005 dollars per second (2026-10-31): Form at listing: an image-to-video model from The Biological Computing Company, live on OpenRouter from October 30. Takes a starting image and prompt to produce 3-second or 5-second clips at 480p or 720p. Billed per output second from 0.005 dollars, about 0.025 dollars for 5 seconds, ranking first on both cost and speed among the 27 video models on OpenRouter with roughly 13 seconds of end-to-end latency. The vendor calls it the first neuron-derived video model; technical details are unpublished.