Skip to content
Bud's Wiki

Alex Ksendzovsky — Mischief 1 biocompute image-to-video model launch

TL;DR

Section titled “TL;DR”

Alex Ksendzovsky (CEO, BioComputingCo) announces Mischief 1, an image-to-video model marketed as the output of four years of work “brewing real brain cells on electrodes to understand how they compute.” The launch tweet claims it is at least 5× faster than its base model and the cheapest I2V model on fal and OpenRouter from $0.005/second, with an AWS marketplace listing to follow. Creative collaboration is credited to Wonder Studios; investor list includes Jeff Dean, Scott Belsky, Thomas Dohmke, Naveen Rao. No model card, architecture details, or benchmarks are disclosed in the thread — the claim is a commercial launch, not a technical report.

Key claims

Section titled “Key claims”
  • Mischief 1 is a biocompute-derived image-to-video model built on four years of research on cultured neuron-on-electrode compute [tweet 1].
  • Pricing is $0.005/second, pitched as the cheapest I2V on fal and OpenRouter at launch [tweet 1].
  • Mischief 1 is “at least 5× faster than its base model” [tweet 1] — the base is unnamed in the thread.
  • Available immediately on fal and OpenRouter; AWS marketplace listing in progress [tweet 1, fal reply].

Method

Section titled “Method”

Not disclosed. The announcement frames Mischief 1 as a commercial product derived from cultured-neuron compute research; no model card, no technical report, no linked arxiv paper, and no architecture or training recipe are provided. The 5× speedup over an unnamed base and the sub-cent-per-second price are the only concrete numbers given.

Results

Section titled “Results”

Results claimed in the launch tweet: (a) $0.005/second price point advertised as cheapest I2V on fal and OpenRouter; (b) 5× speedup over the (undisclosed) base. No quality benchmarks, no VBench numbers, no sample-count or human-eval reports. The attached preview video (~44 s) is the only visible output artifact and could not be inspected from text.

Why it’s interesting

Section titled “Why it’s interesting”

The latency / cost claim (“at least 5× faster than its base, cheapest I2V on fal and OpenRouter at $0.005/s”) sits in the same product neighborhood as the realtime / streaming I2V stack filed under Diffusion serving optimization and the few-step distillation line in Diffusion Distillation — but Mischief 1 is a closed-weights product announcement, not a technical disclosure, so the comparison can only be made against public endpoints. The “biocompute” framing is the novel axis: no filed paper on the wiki describes a production I2V model whose compute substrate is cultured neurons rather than silicon, so until a technical report lands this is a datapoint for whether unconventional hardware substrates are reaching commercial viability for generative video — not for whether the model is architecturally novel.

See also

Section titled “See also”