Spark

Coming to this surface

Love a Prism but need it for your own scenario? Spark is where a scenario becomes a model. Bring your footage, label what matters, and Spark trains a small, edge-deployable model that carries frontier-grade accuracy for your case — without the annotate-train-deploy grind. Think Edge Impulse for video, or Fireworks fine-tuning, but the output is a tiny model that runs in milliseconds where your data lives.

• Upload footage and label the events that matter for your operation.
• Train a small, scenario-tuned model — frontier accuracy, edge latency.
• Evaluate against the frontier model on the Benchmark surface.
• Ship your model as a new Prism, and run it live in Lumin.