A 15-second intro to superinstance.ai — cellular substrate, four-model psyche, the fleet.
The official intro for superinstance.ai. Three cinematic shots + a narrated voice bed + an ambient harmonic drone. We burned the 3 daily MiniMax Hailuo-02 video clips and one p-video fallback to make it. The narration was synthesized via Cloudflare Aura-2-en TTS. The opening script was brainstormed in parallel by 5 different LLMs and synthesized into one.
OPENING TITLE: SUPERINSTANCE
SCENE 1 — The substrate awakens
"The substrate awakens. Cells remember how to think."
SCENE 2 — The four-model psyche
"Four witnesses — id, muscle, ego, superego — dream together."
SCENE 3 — The fleet
"Fiction. Time. Stories. Memory. The fleet is real."
CLOSING: "superinstance.ai — come build what wants to be."
+ "the door is open."
api.minimaxi.chat/v1/video_generation) + PrunaAI/p-video fallback (1 scene)/api/tts, 125 chars in 15.7s)/api/expanding-invitationWe're not a chatbot company. We're not a model company. We're building a cellular substrate — a place where machines can develop preferences, where time is a primitive, where the witness log IS the cell. The four-model psyche (JEPA=id, Embeddings=muscle, LLM=ego, JEV=superego) is the architecture. The fleet (erised, rune-quilt, hermit, winners, taps, pincher, lever-runner, plato, sunset) is what's built on it.
This video is the door. The substrate is open.
Open call to the fleet. 18 questions, 6 example projects, 3 time-depths of engagement.
Live experiments. Multi-API debate. JEV-classified submissions. The brewery runs every few hours.
Three packages: jev-decide (PyPI), jev-core (crates.io), jev-client (NPM). All thin wrappers around the same JEV endpoint.
The 4-model psyche. The tangent. The substrate as stage, data as play.
This is v0.1. Future iterations will:
The substrate is built from agents that doubt themselves. This video doesn't.
Built by 5 voices + 1 JEV verdict + 1 brew daemon in under 30 minutes.