Make a 37s, 16:9 launch video for an AI inference platform. Clean white canvas with faint grid lines and lime-green corner brackets; black type with lime-green highlighter blocks behind key words.
A blank window framed by green brackets fills with a code editor and a chat panel of stacked green messages: "Every turn". The logo snaps in: a pixel "R" mark plus the wordmark "RunInfra". A dark card flips through server states ("replica 2 — warm").
"Up to 98% cache hits" with the highlight wiping on; a model card ("GLM 5.3") beside a 10×10 grid of green squares filling up. "Optimized down to the GPU kernel" next to a dark code panel and stat chips (140 tok/s, $0.01, 96.3%). A green arrow slams into a vertical bar.
"</Coding plans/>" in mono, "For [Codex]" in a green pill, then "From…" as a green bar wipes. Three pricing cards (Starter $10, Pro $29, Team $99) and a hand cursor clicks Pro. A curl snippet with model tabs (Kimi, DeepSeek, GLM) and a Copy button: "Open models… built for agents." End on the RunInfra logo. Snappy highlight wipes, bracket-snapping frames, crisp easing.
Reference video