# Nessco launch film 02 — kinetic edition

The second film at `/animate?film=2` is a 176-beat, 66-second cut at 160 BPM. The original film remains under Launch 01. Tempo can be adjusted from 120–180 BPM, and both landscape and portrait compositions are supported.

## The edit

The opening lasts 8 beats, Nora 32, Physical AI 48, Vision 24, Innovate AI 56, and the closing 8. Innovate AI has four equally timed design sequences: mechanical CAD, PCB, IC layout, and architecture. The longer design chapter shows a condensed design brief, construction steps, and final geometry. Timed stage reveals and exploded views are presentation animation of the returned design documents.

The motion uses rapid angled shutters, perspective pushes, chromatic type entrances, and transparent 3D objects. Every transition derives from the absolute timeline, so seeking backwards and offline export produce the same frames. The exact Nora-Voice Agent title remains visible above the spoken exchange.

## Product and data sources

- Mechanical, PCB, IC and architectural examples: actual authorized Kimi requests through Nessco, saved as editable documents, validated and rebuilt with Nessco’s geometry tools. The PCB and building then receive a documented native detail pass: a populated controller board, a dense framed-window grid and a pitched roof. The original Kimi documents are retained separately from the editable detail documents and roof assemblies. Source prompts, derivations and hashes are in `assets/launch2/engineering-recording.json`. IC and PCB views are layout demonstrations, not fabrication sign-off.
- Physical AI: recorded Go2, H1 and A1 walking policies, Robotaxi vehicle motion, and a dual-arm task-space IK/controller sequence using the actual product controllers and model geometry. The arm grasp uses the product’s assisted kinematic attachment and gravity release. The Robotaxi visuals are shared with the upgraded actual simulator. No new physical hardware training result is claimed.
- Nora: an audible prerecorded exchange with synchronized text and a waveform derived from the audio samples. The audio manifest identifies the source of the reply and the speech voice. See `launch-vision-voice.md` for exact provenance.
- Vision: public-domain real RGB-D reconstructed object scans, with geometric orientation and bounds calculated from their vertices. These are real captured objects arranged for the film, not a claim of a live camera feed or SpatialLM inference. The available SpatialLM examples had noncommercial restrictions and are not redistributed here. See the spatial asset source notes for attribution and calculations.

## Playback and export

Nora voice is included by default and can be turned off under Make it your cut. Adding local music mixes it with the spoken exchange and lowers it during speech. User music stays on the device. Voice, text, and visualization track the selected tempo; pause and seek reposition the same recorded audio.

Video exports render every frame at its exact timestamp through WebCodecs and the pinned Mediabunny muxer. MP4/AVC/AAC is preferred with WebM/VP9/Opus fallback. Both include Nora audio when enabled, even without music. Full HD export, 30/60 fps, PNG frames, and caption downloads remain available. Lower performance devices may render slower than real time without dropping timeline frames.

Reduced motion prevents autoplay; hidden or offscreen pages pause playback. Export cancellation releases encoding resources and returns to the preview. The first film retains its own playback controls.