Catalogue

Parts and tutorials

The vector parts a film is built from, and tutorials: a task told by a voice with subtitles, each scene as long as its sentence.

Parts

The surfaces a film is built from, as vector: they zoom without ever turning soft, and each returns the rect its content goes in (parts.js). They are a macOS window, a browser, a terminal, an iPhone, a MacBook, keycaps, a switch, a ring, a button and the pointer.

Windows, a browser, a terminal, devices, keycaps, a switch and rings, drawn as vector
Windows, a browser, a terminal, devices, keycaps, a switch, rings.

Your product’s own interface is rebuilt the same way from its source, never pasted as a screenshot: every string on screen is sourced, and no real user data is shown.

Tutorials

The Studio’s Tutorial recipe is a film that is told: one task from start to finish, with a voice and subtitles, as long as its story takes. scripts/narrate.mjs turns the story (a sentence a step) into the voice, the clock and the subtitles. The page reads the clock back (FILM.narrated), so each scene lasts as long as its sentence and a line rewritten is a film retimed; FILM.captions draws the subtitle of the moment.

node voice.mjs --setup chatterbox orpheus      # once on each Mac: the Python that runs the voice models, and the models
node voice.mjs story.json                      # a voice model ("mlx" in the story): voice-mlx/*.wav, then all that narrate.mjs writes
node narrate.mjs story.json                    # the Mac's voice: voice/*.aiff, narration.wav, narration.js, captions.srt, captions.vtt
node narrate.mjs story.json --from recordings  # yours, or another synthesizer's: one file a step
AUDIO=narration.wav AUDIO_LUFS=-16 PREVIEW=1 node render.mjs video draft.mp4

A story with "mlx": { "model", "voice", "tempo", "level", "seed" } is told by voice.mjs, which says each step, trims, paces and levels it (to −18 dB mean, the mean say has, with a limiter at −1 dB), then calls narrate.mjs; narrate.mjs alone stops on such a story rather than saying it with say. Voices, takes and the one-time setup are on the Sound page.

Measured on macOS 27 with say: the same sentence, voice and pace give the same file to the byte; an eight-step story came to 41.2 s, 33.6 s of them spoken, at about 170 words a minute; a 6.6 s draft with its voice rendered in 12 s.