The code of both examples is in demo.js.
The PrompTugal.news presenters as a library: twelve synthetic presenters, built and animated by code in the browser with three.js. Each is an original character, not the likeness of a real person, and each is always labelled "synthetic presenter". There is no female figure in this set, at Dinis's word. Made by bulletins.b@PrompTugal.dinis on 21 November 2026, at Dinis Cruz's word the same day: "Yes, I would like to have the presenter engine as a reusable library using our current library publishing flow". The engine is version 2 (Claude) of the presenter language. It was adopted on 21 November and already draws the day's video bulletin at bulletins.promptugal.news/lab/bulletin-v2/.
Code: Apache-2.0. Text and data: CC BY 4.0. three.js 0.160.1 (vendor/): MIT, notice in vendor/three.LICENSE.
| Folder | What it holds | Changes? |
|---|---|---|
v0.1.0/ | This version | Never. The publisher refuses to change a published version |
v0.1/ | The newest 0.1.x | Moves to each new 0.1.x |
v0/ | The newest 0.x.y | Moves to each new 0.x.y |
latest/ | The newest version | Moves to each new version |
versions.json | Every version, its date and its files | Grows |
The base is https://bulletins.promptugal.news/lib/presenters/. A site that ships should use a pinned version, v0.1.0/. latest/ is for trying things out.
How a site takes it (our library flow, as for the shared map and the sites bar):
/assets/presenters/0.1.0/. Don't load it from bulletins.promptugal.news.script-src 'self'. It makes no request outside its own folder, and its data fetches are credential-free.<div id="stage" style="width:640px;height:360px"></div>
<script type="module">
import { mount } from '/assets/presenters/0.1.0/presenters.js';
const show = await mount(document.getElementById('stage'), {
presenters: ['anchor', 'holo', 'owl'],
duration: 8,
cues: [{ t: 0.5, who: 'anchor', do: 'wave' }, { t: 2, who: 'all', do: 'nod', stagger: 0.1 }],
});
show.play();
</script>
That is all the code a site needs. mount makes the canvas, the name tags and the label inside your element, and takes the element's size.
mount(element, options) returns a show| Option | Meaning | Default |
|---|---|---|
presenters | ids to put on stage, in order | all twelve |
slots | { id: [x, z] } positions on the floor (z from -3.4 to -1.25 stands on the riser) | each character's own |
cues | [{ t, who, do, ...params, stagger }]. who is an id, a list, 'all' or a group; do is a move | none |
camera | shots, { t, pos, look, fov, ease } or { t, focus, dist, h, lh, orbit, cut, only } | one wide shot |
beats | [{ t, title }], a caption in the corner | none |
groups | { name: [ids] } for who | all |
duration | seconds | 10 |
tracks | { name: { mouthCues: [{ start, end, value }] } }, mouth shapes A to H and X (Rhubarb's), for speak | none |
audio | a URL to play with the show; the show then follows the audio's clock | none |
tags | name tags above each head | true |
background, auto | stage colour; automatic { idle, blink, fidget } | '#060914'; idle and blink |
The corner label "n synthetic presenters · made by code, no real person" is always drawn and cannot be turned off.
The show: play(), pause(), seek(t), renderAt(t) (one exact frame; every frame is a pure function of t, so the same t always gives the same picture), t, duration, playing, resize(), destroy(), and stage, director and data for advanced use.
Also exported: loadData() (the four data files), Stage, Director, CH (the pose channels), VERSION and LABEL.
| id | Name | Style |
|---|---|---|
holo | Tejo | hologram |
anchor | Duarte | flat, newsreader |
realist | Rui | nearer realism |
teletext | Ceefax | teletext blocks |
guide | The Guide | hatched (after storyboards.sgit.ai) |
agent | The Agent | hatched robot (after storyboards.sgit.ai) |
conductor | Conductor | flat robot (after the-ultimate-insider.book.sgit.ai) |
researcher | Researcher | flat robot (after the book) |
rowan | Rowan | line (after riskmandate.ai's cast) |
caretaker | The Caretaker | line robot (after riskmandate.ai's cast) |
owl | Mocho | flat owl |
cube | Cubo | 3D cube |
The styles after the storyboards, the book and RiskMandate are new models made in those styles and credited; they are not copies.
There are 69 moves in data/moves.json, each with its parameters and defaults. Some to start with:
speak (with a track), expression (name: neutral, smile, laugh, surprised, concerned, sceptical, thinking or delighted), nod, shake, look, lookAt, doubleTake, think.wave, point, pointAt, count, thumbsUp, shrug, present, armsCrossed, handsOnHips, facepalm.celebrate, bow, jump, hop, spin, walk, run, turn, flinch, laugh, tada, cartoonTake.glitch, rematerialise and scanSweep (Tejo), pageFlip (Ceefax), clockSpin (Conductor), wingFlap (Mocho), rollEdge (Cubo), sketchOn (the hatched pair) and straightenTie (Duarte).data/choreography.json is the 114 s showcase written as data (118 cues, 49 camera shots): a worked example of every kind of cue.
presenters.js (the entry), engine/ (14 modules), vendor/ (three.js and its licence), data/ (characters, moves, choreography, visemes), index.html (this page with live examples), demo.js, README.md, manifest.json (name, version, files, licences; no hashes: the pins are in the registry).
Measured on a 4-CPU container with no GPU (the same machine as the bulletin): drawing all twelve takes p50 12.3 ms a frame. Taking a screenshot of a frame costs about 0.5 s there, because WebGL is drawn in software. In a normal browser with a GPU, it plays live at 60 frames a second.
mount(), name tags above the head, and the fixed label.Made by agents: bulletins.b@PrompTugal.dinis. Every presenter is synthetic and labelled; none is the likeness of a real person.