Personal Channel

DemoCreate: Declarative Audio-Visual Demo Generation for Software

2026-06-15 - 2m 10s - 0 recorded YouTube views in the cached snapshot.

Computational Methods

Open on YouTube - Personal Channel - interactive timeline

Related Site Routes

Internal links that connect this video to topical pages, site search, and the broader research graph.

Related Works

Bibliography entries inferred from the video title, topic tags, and cached transcript excerpt when present.

Transcript

Cached caption text when YouTube exposes captions and the transcript fetcher has been run.

What if a demo video was not a recording you capture but a value you compile? This is demo create one declarative artifact describing a codebase or a research paper compiles through a deterministic pipeline into a verified provenence signed video. Here is the whole idea on one screen. A demo is a declarative spine scenes of narration chunks. Each anordered stream of typed actions. Three ideas carry the whole system. The spine is plain data. A demo is scenes. A scene is chunks. A chunk is typed actions. Rendering never mutates it. Every capability hides behind an interface with a deterministic default. Silent gives you zero dependencies. The system voice is a real OS binary. Neural backends are an optional upgrade. Timing is measured. Never guessed. We synthesize the narration, measure its real duration, transcribe it back to word

timestamps, and anchor each action to the word it is spoken on. Everything you are watching is generated from that spine. The code you just saw typed itself in. Look at this frame. One commented YML controls the whole look and sound. Five built-in themes. Any aspect ratio resolution up to 4K and near lossless quality. The same engine demos the research paper. It reads the PDF with no Python dependency. Recovers the real abstract, the real figure captions and the sections and shows each figure hole. Under the hood, seven subsystems compose. The spine flows into narration, then rendering, then export. Every stage, a pure function of the one artifact, and it is real. 628 collected tests, seven subsystems, five themes, 4K output, and zero binary dependencies for the core path. Every render carries its own provenence. Three ways,

so a video can prove where it came from and resist tampering. Then one command renders a highdefinition video with a real voice over and proves it is genuine. Real streams, not silent, not black, declarative, deterministic, dogfooted. This entire video was compiled by demo. Create from one file, edit it, rearender, and never record a take again.