Invisible Network
A Natural History of Planetary Intelligence vs. Human Stupidity
TV. Music. Books. Games. News. One platform. Built to prove a point.
What This Is
Invisible Network is a technology demonstration and companion platform for the book Are You Smarter Than Earth? by David Jean Charlot, PhD. It is not a commercial media network. It is a proof of concept — a working prototype that demonstrates how intelligent systems can be designed with energy-first principles, distributed architecture, and regenerative thinking.
The book asks whether humans are smarter than the planet they live on. This platform is part of the answer: technology that embodies the lessons Earth has been teaching for 4.5 billion years — efficiency, adaptation, feedback responsiveness, and zero waste.
Who Makes It
Invisible Network is built on Invisible Infrastructure, created by David Jean Charlot, PhD as part of his book Eco AI: The Economics, Ecology, and Ecosystem of Artificial Intelligence (The Human Condition, Vol. 2), from his Fabric library of non-fiction.
Invisible Network and Invisible Infrastructure are projects of Voyage Tracable, LLC, David's creative holding company. David's own website is davidjcharlot.com.
The site's energy figures follow the energy measurement of WAI (Web AI Media Transport & Execution), the open media standard of Transaction Science, Inc.: each figure carries how it is known (its acquisition class) and how far it can be leaned on.
Than Earth?
Are You Smarter Than Earth?
A Natural History of Planetary Intelligence vs. Human Stupidity
Earth figured out how to regulate its own temperature for 4.5 billion years. You can't even regulate your thermostat without an argument. Earth built a self-sustaining ecosystem with millions of interdependent species. You can't keep a houseplant alive.
This book traces Earth's intelligence across billions of years, then compares it to humanity's 250-year industrial experiment. The results are not flattering.
Eco AI
The Economics, Ecology, and Ecosystem of Artificial Intelligence
The economic and ecological dynamics of AI, past, present and future. Invisible Infrastructure, on which this site is built, is part of this book.
How much energy and data an hour of media takes, stage by stage, with levers to move and what this site's own servers sent: Telemetry.
AI Artificial Intelligence Disclosure
What this site uses machine learning for, where each runs, and how you can tell; then what it uses that might be taken for machine learning and is not, and what it does not do. Each line is checked against the code that does it.
Subtitles transcribed by machine
Where one of the site's own films came without subtitles, what is said in it was transcribed with Whisper large-v3-turbo, OpenAI's speech-recognition model, whose weights OpenAI publishes under the MIT licence, on our own computer before the subtitles were published. They are kept as files; nothing is transcribed while you watch. Subtitles people wrote, such as Wikimedia Commons contributors' or a film's own, are used instead wherever there are any.
The film's transcript page says “Machine transcription by Whisper, not reviewed by a person”; subtitles people wrote are credited on the title's page.
Audio descriptions written with AI assistance
The audio descriptions of the site's films, which say what the picture shows for viewers who cannot see it, are each written with AI assistance: by Claude (Anthropic), from frames of the film, and timed to it. Your device speaks them in its own voice (the browser's speech synthesis), or shows them as text.
The title's page and its transcript say “written with AI assistance”, and each description file carries Content Credentials (C2PA) that say, signed and in a form machines read, that it was made by an AI model.
Chapter names written with AI assistance
The chapters of the site's own films start at the cuts, fades, black and silences FFmpeg measures, moved to where the story turns; their names were written with AI assistance, from the frames, from what is said, or from the intertitles.
Each film's chapter file says how its chapters were made.
Channel lines written with AI assistance
The lines that say what is on each of the site's genre channels, in each channel's own voice, were written with AI assistance (Claude, Anthropic): sentences filled from each title's record (its name, length and year) and from the schedule (the time, the daypart, the set piece, a season's night), and a sentence more for a handful of titles, each checked against the page it cites. They say nothing their record or the schedule cannot back. Nothing is written while you watch, and no model runs on the site for them.
Each channel's page and Tonight say so beside the lines.
Stories to play, written with AI assistance
The stories to play under Games are written with AI assistance (Claude, Anthropic) from the text of a public-domain book, each from a world file of the book's characters, places and facts. Every line is tied to the passage of the book it comes from, and the story's page shows that passage, and links into the reader at it. A story is written once and kept as a file: every visitor reads the same words, and nothing is written while you play. No model runs on the site for them.
A story's page says so under its title, with whether a person has yet reviewed it, and Games marks each one.
Sharper pictures, on your device
Low-resolution films can be upscaled as they play, on your device's graphics chip. Drawn animation uses Anime4K's CNN, a small convolutional neural network trained by its author (bloc97, MIT licence), run as WebGPU passes or, where the browser has it, as a WebNN graph; live action uses AMD FidelityFX Super Resolution 1, a fixed filter that is not machine learning. The detail it adds is an estimate, not the film's own. No frame leaves your device.
The player's settings menu shows it, and turns it off.
Follow my head: a face detector on your device
On the wall, Follow my head turns your computer's camera on, only when you press it and your browser asks you, and a small face detector (BlazeFace, a convolutional network trained by Google, Apache 2.0 licence) finds where your face is in each frame, in your browser, on your graphics chip where it can and otherwise on its processor. Only where your face is moves the view; no picture is shown, kept or sent, nothing about your face is kept, and the camera stops when you stop following or leave the page.
The button says what the view follows, and the browser shows that the camera is on.
Window mode: your eyes found on your device
On the wall, and on a work's page or the gallery under Art, Window mode turns your computer's camera on, only when you press it and your browser asks you. MediaPipe's Face Landmarker (neural networks trained by Google that find a face and the points of its mesh, the irises among them; Apache 2.0 licence), run by the face tracker this site shares with NeuroPlay, finds where your eyes are in each frame and, from the size of your irises, how far they are from the camera, in your browser, in a worker on your graphics chip (where it cannot run there, the pointer moves the view instead). Only that position moves the view, drawn as through a window; no picture is shown, kept or sent, nothing about your face or eyes is kept, and the camera stops when you turn it off or leave the page. On a phone or tablet it follows the device's tilt instead, with no camera.
A line beside the button says what the camera is for, the button says how to stop it, and the browser shows that the camera is on.
The body game: your arms found on your device
On a sport's page with a body game (archery's), Use my camera turns your camera on, only when you press it and your browser asks you. MediaPipe's Face and Hand Landmarkers (neural networks trained by Google that find a face and the points of a hand; Apache 2.0 licence), run by the tracker this site shares with NeuroPlay, find where your eyes, your bow arm and your draw hand are in each frame, in your browser, in a worker on your graphics chip. Only those positions draw, aim and release the arrow, and only the last moments of them are kept, for the shot; no picture is shown, kept or sent. The game plays as well without the camera.
A line under the button says what the camera follows and that nothing is sent or kept, and the browser shows that the camera is on.
Translated captions, on your device
Where your browser has a built-in Translator (Chrome's, which runs a translation model Google makes and the browser keeps on your device), a film's captions can be translated into your language. Nothing is sent to us or to anyone else.
The subtitle menu says “translated on this device”.
Asking aloud
The microphone in the search field uses your browser's own speech recognition: on your device where the browser can, otherwise at the browser maker's servers, which the page tells you first, waiting for you to agree. Only the words reach this site, never a recording.
Before you speak, the page says where your speech will go.
Books read aloud by your device's own voice
In the book reader, Read aloud can have your device read the book's own words in its own voice (the browser's speech synthesis): only the voices your browser says it runs on the device are listed, each word is lit as the voice says it, and nothing is sent to this site. The voice is your device's, as a screen reader's is; this site makes no sound for it.
The panel names the voice and says it is your device's before it reads, and the bar while it reads.
Machine-made text from the libraries
Some text comes from the sources already made by machine: Project Gutenberg's summaries of its books, the Internet Archive's automatic transcripts of radio drama and subtitles of films, the subtitles on Wikimedia Commons that their page's record or their own text shows a machine made (speech recognition, automatic captions, a machine translation), and the text the Library of Congress and the Internet Archive read from scanned pages (OCR). Where neither the record nor the text says who made a film's subtitles from Commons, the page says the maker is not recorded.
Each says so where it is shown.
A voice in the studio, generated on your device
In the studio, once you have put something of your own in, you may ask a model to read your own words aloud: Piper's en_US-ljspeech-medium, a text-to-speech network trained from scratch by Bryce Beattie on the LJ Speech dataset, Linda Johnson's public-domain readings for LibriVox, and released into the public domain. It runs in your browser, on your graphics processor through WebGPU where the browser gives one, else on its processor, through ONNX Runtime Web; eSpeak NG's hand-written rules turn your words into the sounds it reads. Each time it makes three takes that differ, along named dimensions you set (pace, expression, rhythm), and you choose one or none. Its files come from this site and are checked before use; your words stay on your device. It is not offered on a child's profile, and not in China or India (/legal says why).
Every take is labelled “AI-generated” beside it, on its tile, on the picture when it is first heard, on the card a recording ends with and in its credits file, which also names the model and, where its maker says, what it learned from; a recording that uses one is labelled AI-assisted, and its picture carries the EU's “AI MODIFIED” icon top right while the voice is heard, at its start, every half minute and after a break. The take's file, and an MP4 recording or edit that uses one, carry XMP metadata that says so to machines (IPTC digital source type). Everything it says carries a watermark in the sound itself, and a take's file and such an MP4 a manifest signed in your browser with a key made there (the MP4's names the take in it), which says the file has not changed since and nothing about who you are; anyone can check a sound or a recording for them on /studio/check, on their own device.
The Intelligence station's briefing, read aloud on your device
On the Intelligence station you may ask for the day's briefing to be read aloud: the studio's voice, Piper's en_US-ljspeech-medium (trained from scratch on public-domain recordings and released into the public domain), runs in your browser and reads the briefing's own words, nothing else. Its files come from this site and are checked before use; nothing leaves your device, and nothing is kept. It is not offered on a child's profile, and not in China or India.
The button and the note beside it say the voice is AI-generated and name the model, before it reads and while it reads; everything it says carries the studio's watermark in the sound itself.
Books read aloud in the reader by an AI-generated voice, on your device
In the book reader you may choose the studio's voice to read an English book aloud: Piper's en_US-ljspeech-medium (trained from scratch on public-domain recordings and released into the public domain) runs in your browser and reads the book's own words, a sentence at a time, nothing else. Its files come from this site and are checked before use; nothing leaves your device, and nothing is kept. It is not offered on a child's profile, and not in China or India.
The voice is listed as AI-generated with its model's name, the note beside it says so before it reads, and the bar says so while it reads; everything it says carries the studio's watermark in the sound itself.
Read along: recordings timed word by word, on our own computer
Where a LibriVox recording of a Project Gutenberg book has been timed, the reader can play it with each word lit as the volunteer says it. The times were found once on our own computer by wav2vec 2.0 (Large, LV-60 with self-training; Meta, Apache 2.0 licence, the model the subtitles use): it heard each chapter, the runs of words it heard as the book writes them were held, and the book's words between them were fitted to the sound (CTC forced alignment). Only the times are kept, beside each word of the book; the voices are the volunteers' own, unchanged. No person has checked the times; where a chapter of the recording is not found in the text, it says so and is not timed.
The reader's panel and the recording's page say “words timed by machine (wav2vec 2.0), not checked by a person”, and how many of the recording's sections were timed.
AI Future's genre maker: a channel from your words, chosen and ordered by a language model
In AI Future, the part of the site behind a door for people aged 18 or over, a signed-in person can type a few words for a channel of their own. The site's rules first find the titles its channels air that fit the words, for an adult's profile where the person is. Then, where it runs, a language model, Gemma 4 12B (Google's open-weights model, Apache 2.0 licence), on our own servers, chooses and orders some of them and writes the channel's name and a one-line description, held by a grammar to the titles it was given and to the catalogues' own words; the server checks every line again. It never adds a title the rules did not find, says nothing about the person, and nothing about how good a work is. The words and the titles go to the model on our server and nowhere else; the channel, its words and its titles are kept in the person's account until they delete them. Where the model does not run, or what it writes does not hold, the rules' order and words stand, and the channel says so. A channel is refreshed later by its rules alone.
Each channel carries its label, AI-curated or Curated by rules, with what it means and what making it took, by fabric; what the model chose and wrote is marked for the human-made-only view.
AI Future's game maker: a body game of a sport's move, shaped by a language model, decided by the site's engine
In AI Future, the part of the site behind a door for people aged 18 or over, a signed-in person can make a body game of a sport's core move: a serve, a penalty save, a judo stance. The site's rules make it first, from the sport's own rules and measured numbers, each with its source. Then, where it runs, a language model, Gemma 4 12B (Google's open-weights model, Apache 2.0 licence), on our own servers, may shape it from the person's words: it picks one of the sport's moves, names it with the sport's own words, chooses its how-to lines from the sport's own guide, word for word, and sets the few numbers the move allows, held by a grammar to that form. It writes no rules, no physics and no free text. The site's engine then checks the game in stages (its schema, the sport's cited ranges, its physics, its safety in a room, and play by a simulated player at every level); a game that fails is refused with the reasons, which go back to the model, and after ten answers at most the rules' game stands. A game may also be made from a moment the person just watched on the site (Play this moment): only what the site's own records state of it is read (the work's name, its chapter, where in it, whom it is from), never its picture or sound; the facts it states become the game's numbers where the rules have a cited number for them, the model is never given them, and the game names the moment it is from and says it is the site's or AI-made, never the broadcaster's. The words, the sport's data and, for a moment, the work's name and chapter go to the model on our server and nowhere else; of a video on YouTube nothing is given to it or read from its title. Nothing is used to train anything.
Each game carries its label, AI-made or Made by rules, with what the engine checked, the model's answers it refused, its sources and what making it took, by fabric; what the model shaped is marked for the human-made-only view.
Book Illustrations
Are You Smarter Than Earth? contains AI-generated scientific illustrations produced in a naturalist-bioluminescent style. These images are clearly identified as AI-generated within the publication.
Not machine learning
“Catch me up”: lines from the film itself
For a film you left part-way through, “Catch me up” shows lines said in it before where you stopped, as its subtitles give them, each with its time and opening the film there: one from each part of what you watched (where a new voice comes in, or after a pause, if there is one) and the last line before you stopped. Nothing is written, summarised or generated, by us or by your browser; it is worked out in your browser and nothing is sent.
Under the list, the page says how its lines were chosen.
Clips: cut and encoded in your browser
A clip of up to a minute of a film whose licence allows it is cut in your browser and encoded by your device's own video encoder (its media engine where it has one, else its processor), with the film's title, credit, licence and a link back in the file. Subtitles go into the picture only where you choose it and the licence allows it. Nothing is generated, and no model is involved; nothing of it reaches this site, and nothing is kept.
Beside the clip, the page says what it carries and which part of your device encoded it.
The studio: mixed and recorded in your browser
In the studio, what you mix (your camera, your screen, your files, a phone of yours, guests' cameras, and works from the library whose licences allow it) is drawn and switched in your browser, and recorded there by the encoder the browser chooses (your device's media engine where it uses one, else its processor). The key takes out a colour you choose, by arithmetic on each pixel's colour on your device's graphics processor: it is not a model that finds a person. A phone's picture and sound come straight to your browser over your own network and never reach this site; only the details the two devices need to find each other pass through it, for a few minutes, and are not kept. A guest's picture and sound come straight to your browser too, and never reach this site, and they get the program back; the name they type goes to you alone. Going live sends the program straight from your browser to an address you name, by WHIP, with a token that stays in your page's memory: this site hosts no stream, and sees neither. Its editor cuts your recordings, your files and those works together on a timeline in your browser, by the rules you set (trims, splits, moves, mixes, titles), and makes them into one file with your device's own video encoder, with every work's credits in the file. Nothing in the switcher, the key, the recording or the editor is generated, and no model is involved in them (the voice, where the studio offers one, is its own entry); nothing you make reaches this site, nothing played there is counted, and nothing is kept unless you press to keep an edit, which stays in your browser until you delete it.
The studio's page says so under its title, the phone's page under its own, and a guest's page says who sees them before the press; beside a recording, and beside the editor's file, it says on what terms it may be shared and which part of your device encoded it.
Your rows, and what Moments learns
Home's “Continue watching” and “Because you watched” rows, and what Moments learns from what you stay with or skip, are worked out in your browser from what it keeps, by counting and fixed rules (in Moments, each stay weighed by the chance a moment as new to you had of being drawn, and by where it came in its set of 20). None of it is sent to us. In Moments, one press plays a set of up to 20 moments in a row, then the set stops; you can choose instead to have each moment wait for your press, and “Less of this” on any moment leaves its kind out, with or without personalisation.
Settings shows what is kept, and forgets it; each moment says why it was chosen.
Your channel, in Yours
Your channel in Yours plays the genre channels' titles in an order worked out in your browser, by counting and a fixed rule, from what this browser keeps: your watch history, My List, the diary and ratings you keep in Yours, and, where Moments' personalisation is on, what Moments learned. Four titles in five come from the channels in proportion to how much you watch, log and rate from each (a smooth weighted round robin); one in five from the channel it has shown least so far, so it does not narrow; nothing you finished or logged comes back. None of it is sent to us: this site is asked only for the channels' titles your profile may watch, the same for everyone with it, and for each title's file as it comes up. One press plays the title picked first and each next one at its start, until you stop it or leave the page; after four hours with no press it stops at the next title and asks.
Under each title, the page says why it was picked; Settings shows what Yours keeps.
One for tonight
On Home and TV, one title, channel or marathon is picked for tonight in your browser, by counting and a fixed rule, from what it keeps: what you played and how far, the genres each title is filed in, your places in marathons, the site's channels you starred in the guide, and what you said Not this again to. Each genre, radio series and kind of medium gets a vote from every title finished, half of one from every title watched a while, and loses half of one for every title left in its first minutes; votes halve each week. A channel weighs its genre's votes, its series', half its medium's, and two more where it is a favourite; its title on now is worth that times the share of it left, a title later tonight a little less the later it begins; a title left part-way in the last fortnight, or a marathon's place, is worth more, and less each day. The most worth is the pick, said with its reason; nothing finished or said Not this again to comes back, and where nothing leans anywhere it is what is on now, the same for everyone. Buttons for mood, company and time narrow what it picks from, by fixed rules and not by a model: a mood keeps to the genres it names (Something light to comedy and animation, Something gripping to mystery and crime, film noir and adventure, Something scary to horror), With kids to what is placed for children, On my own counts what you left part-way double, With friends leaves it out, Under 30 minutes keeps to what has that or less to go, and A whole evening to a channel, a marathon or a title with an hour or more to go; where nothing fits, the time and then the mood are let go and the sentence says so, but With kids never is. They are kept for tonight, until 06:00, and with personalising off for the page only. None of it is sent to us: this site is asked only what the channels air now and tonight and which of the site's own titles play here, the same for everyone with your profile where you are. Its press plays it; nothing plays before.
The pick says why in a sentence; Settings lists every item it goes on and removes each, or all.
The wall: which picture each tile fetches
On the wall, one press starts every tile on it, with sound from one tile only, and nothing on it plays before that press. Which picture each tile fetches is decided in your browser by a fixed rule: no bigger than the tile is drawn on your screen, and for the tiles you are not watching the smallest that is still watchable (240 lines). No learned model is involved, and what is on your wall is kept in your browser: this site is asked only what each tile plays, by its channel's or film's name. A tile in YouTube's player (a film from a rights holder's channel) is not under this rule: YouTube's player chooses its picture.
The wall's page says it under its title, and each tile shows the picture it takes.
Labels, and human-made only
Things on this site are labelled human-made, AI-assisted or AI-generated. A work counts as human-made unless its own record says a model made it, by fixed words in that record; no model looks at a work to decide. What the site or a source added with a model carries its own label beside it. Human-made only, on the settings page, is a choice kept in your browser that leaves out everything labelled otherwise, from a fixed list that says what it covers and what it does not.
Settings lists what it leaves out and what it does not cover, and a line says so where something was left out.
Karaoke: your pitch, measured on your device
On Karaoke, with the microphone on, the pitch you sing is measured in your browser by the YIN method (arithmetic on the sound, de Cheveigné and Kawahara, 2002) and compared with the score's notes there; nothing of it reaches this site. The line under Korean and Japanese words is their romanisation by the official standards' rules, computed on our server from the score's own words. No learned model is involved in either.
Each song's page says so beside the microphone and under the romanised words.
Symbols over words in the book reader
In the book reader you may turn on symbols over words, as augmentative and alternative communication uses them: a picture from Mulberry Symbols (Steve Lee, CC BY-SA 4.0, the site's own copies) above each word the set names, in English and, for some words, Spanish. The words are matched by rules: each symbol's name and its forms (plurals, the forms of verbs, comparisons), never a model; function words and names get none.
The panel credits the set and its licence, and so does a line above the text while they are shown.
Ask the library
A request in plain words is read on our server by rules and small vocabularies drawn from the catalogues' own labels, and its answer is chosen from the catalogues by fixed rules. No language model, and no other service, is involved in either.
The genre channels' hours
Each genre channel's day is laid out in your time zone by rules written by hand: its dayparts, the strands of its titles that fill them, its set pieces at fixed times and its seasons. One day in four a daypart being tried airs another strand; how many tuned in to each and stayed ten minutes is counted per day, with nothing about anyone, and a daypart's strand changes only when Fisher's exact test finds a difference significant at 5% shared over every daypart being tried, from enough tune-ins and sources. No learned model is involved.
TV's line-up and the music stations
The line-up suggested on TV and the order of tracks on the music stations come from rules written by hand (the time of day, tone, pacing, tempo and key); no learned model is involved. TV's line-up holds only what your profile may watch where you are, and the buttons for mood, company and time beside the pick for tonight narrow it in your browser, by each title's genres, age placing and running time.
The Intelligence station's forecasts and briefing
The forecasts on the Intelligence station are made by stated arithmetic from public data: the spread of a series' own past changes over the forecast's horizon, applied to its latest value, or the share of the station's own past daily reads on which something happened. Its daily briefing is put together by fixed rules from its record. No learned model is involved in either, and no person picks or edits a forecast.
Each forecast's page states its method with its numbers, and the briefing says that no model writes it.
What this site does not do
- It trains no models, on visitors or on anything they do. The models above (Whisper's, Claude's, Anime4K's, BlazeFace's, MediaPipe's Face and Hand Landmarkers, Chrome's translation model, Piper's, wav2vec 2.0's, the browser's own) were trained by others and are used as they are.
- It keeps no viewing or listening history on its servers. What you watched, heard, read, stayed with or skipped stays in your browser. The servers keep sums per day with nothing about who: what they sent, how fast pages loaded, how many times each work began to play, how often each of home's rows and the header's sections was chosen on the TVs that keep their usual order, what the players noted about energy (the kind of screen and decoder a video began with, eco mode, videos paused because nobody could see them), how many sets of Moments reached their end and how often Stop here or More was pressed there (with how sets play in that browser, whether it is one of the one in ten chosen at random to start by waiting for presses, and the set's number in the visit), how many tuned in to each genre channel's dayparts and how many of them stayed ten minutes, and the servers' own energy.
- It makes no synthetic media of its own: no generated presenters, faces, pictures, music or video. A voice is generated only on your device, when you ask: in the studio from words you write, on the Intelligence station reading its daily briefing, and in the book reader reading an English book's own words; it is labelled AI-generated wherever it goes. The other synthetic voice is your own device's, reading audio descriptions and books aloud when you ask. It shows films that their own makers say were made with AI, posted on their own channels: each is labelled AI-made wherever it shows, and the site made none of them.
- It sells, places and inserts no advertising; some live channels' streams and YouTube's player carry their owners' own ads. Nothing about you is sold or shared.
If you find something machine-made that does not say so, please tell us.
Content Attribution & Licensing
Books and audiobooks
The Books section reads the catalogues these libraries publish for the purpose, fetches each text once (from Project Gutenberg's mirrors, as its robot policy asks) and keeps it; covers are kept here too, so that browsing does not send you to their servers. Moments and the front page draw passages from a set of books and recordings chosen from the catalogues alone (for each age of profile, an equal share for each of Gutenberg's categories and LibriVox's genres, and within them each language; no reading or download counts) and brought up to date each week as the catalogues grow; each of the site's servers keeps the whole set, taking what it lacks from the other first, so both show the same.
Project Gutenberg (gutenberg.org): its books are in the public domain in the United States; readers elsewhere should check their local laws. Its catalogue data is in the public domain. Every book is shown with the notice its licence asks for and a link to the Project Gutenberg License; its summaries are written by machine, and say so.
Open textbooks (Open textbooks): from OpenStax at Rice University (each book under its own Creative Commons licence, most CC BY-NC-SA 4.0; read at OpenStax or as its PDF, and its covers, which OpenStax reserves, are not shown), LibreTexts (each book under the licence LibreTexts records for it; read here page by page from LibreTexts' own pages), the University of Minnesota's Open Textbook Library (its records CC0; each book under the licence the library states, read at its publisher) and Wikibooks (CC BY-SA 4.0; read here chapter by chapter, each linked to its page and history). Their covers here are drawn.
LibriVox (librivox.org): audiobooks of public-domain books, read by volunteers and placed in the public domain; the recordings are kept by the Internet Archive and play from it.
Standard Ebooks (standardebooks.org): public-domain books corrected and typeset by volunteers, who dedicate their work to the public domain (CC0). This site reads its open New Releases feed; its full catalogue feeds are for its Patrons Circle. An edition's pictures come inside it, and are sent from this site with the text.
Wikisource (wikisource.org): transcriptions under the Creative Commons Attribution-ShareAlike 4.0 licence, each page credited and linked to its history. Shown to teen and adult profiles.
Reading aloud in the reader. Read aloud uses your device's own voice, or, for English books where it is offered, the studio's AI-generated voice (Piper, on your device, labelled so). Read along plays a LibriVox recording with each word lit: the times were found on our own computer by wav2vec 2.0 (Apache 2.0) and no person has checked them. Symbols over words are Mulberry Symbols (mulberrysymbols), © 2018–2026 Steve Lee, under the Creative Commons Attribution-ShareAlike 4.0 licence, served from this site as the set publishes them; their Spanish names are this site's own translation of the set's English ones.
Which profiles see a book is read from what the libraries say about it (subjects, bookshelves, genres); erotica is for adults only. We do not claim copyright over public-domain works.
Music
Music tracks are sourced from the Internet Archive (archive.org) and the Free Music Archive. All tracks are either in the public domain or licensed under Creative Commons (CC BY, CC BY-SA, CC BY-NC). Original artist attribution is displayed for every track.
Podcasts
The shows under Podcasts are listed from Apple Podcasts: its charts in the 180 countries that have them, read once a day (each category every three days), and its directory, searched when you search. Where it is set up, Podcast Index (podcastindex.org), the open directory, is searched too. Every show and episode belongs to its publisher: episodes play in your browser from the publisher's own feed and files, with their chapters and transcripts where the feed has them; this site hosts none of it, and each show links to its website and to Apple Podcasts. The feed of every show in the top 40 of a chart is read about once every two days, and its newest episode fetched from our server the way your browser would; shows that do not answer are kept out of the lists until they do. Which profiles hear a show: explicit shows (by Apple's advisory, Podcast Index or the show's own feed) and shows about sexuality only adults; true crime from 13; children's profiles only shows made for children (Kids & Family, its stories, learning and animals, not parenting); explicit episodes only adults, whatever their show. Shows with adult words in their names are for adults only, and audiobooks are left to the library. Names and artwork belong to their publishers, who do not endorse this site.
Free Music
The music under Free Music comes from the open catalogues, read through their public APIs: the Internet Archive (its Netlabels collection, the Free Music Archive's catalogue as archived there, Jamendo's archived albums, the Live Music Archive of etree.org, including the Grateful Dead's, the Great 78 Project's 78 rpm records, and Musopen's public-domain classical recordings), Wikimedia Commons, and Jamendo where it is set up. Every release names its artist and its licence as the catalogue states it (or says that it states none) and links to its page there; the tracks play from the catalogues' own servers, and this site hosts none of them. Jamendo's music is not offered for offline listening, as its terms ask. ccMixter is not listed: its server plays its tracks only on its own pages. None of these catalogues rates music, so which profiles hear a release is read from its own words: tagged explicit, adult words, swearing or slurs in its title, or the racist caricature some early records carry, only adults; spoken word and comedy from 13; children's profiles hear music made for children, Musopen's classical recordings, and instrumental music. Releases whose files the catalogue no longer serves are kept out of the lists.
Karaoke
The songs to sing on Karaoke are made from public-domain scores: the OpenScore Lieder Corpus (CC0, by OpenScore and its volunteers), the Mutopia Project (each score under the licence Mutopia states for it, credited on its page) and the scores in Japanese Wikipedia's articles (CC BY-SA 4.0, by each article's contributors). Dates of the composers and poets, and when each song was first published, are as IMSLP, the LiederNet Archive, Mutopia, OpenScore and Wikidata state them, each named on the song's page. Piano: the Salamander Grand Piano by Alexander Holm, licensed under Creative Commons Attribution 3.0 (CC BY 3.0); the karaoke channel plays 30 of its samples, shortened, mixed to mono and converted to MP3, from FreePats' edition. Its live channels come from the iptv-org database and its stations from Radio Browser, as on Live TV and Radio.
Radio Drama
The series under Radio Drama come from the archives’ open collections, read through their public APIs: the Internet Archive (its Old Time Radio collection, with the series sets the Old Time Radio Researchers Group has checked and dated, collectors’ uploads, and the Suspense, Whistler and Casey, Crime Photographer projects; WWII Era Audio Recordings; the Pacifica Radio Archives; the Radio Haiti Archive; German radio plays; and what listeners have uploaded there and tagged as radio drama), Wikimedia Commons (historic broadcasts and speeches, among them the Netherlands Institute for Sound and Vision’s Radio Oranje and other Dutch broadcasts of the 1920s to the 1940s), and the Library of Congress. Every episode plays from its archive’s own server, and this site copies none of them. Each set shows its rights as its archive states them (a licence, an uploader’s mark, or a statement) and says whose they are; nothing is called public domain on this site’s word, since under United States law sound recordings first fixed before 15 February 1972 are protected federally by the Music Modernization Act, so most American radio of the 1930s to the 1950s is not in the public domain there. So an episode whose only basis is its uploader’s public-domain mark is not played, and its page says why; an episode under its rights holder’s licence plays everywhere (where each plays). Series, networks and air dates are read from the archives’ own titles, descriptions and file names; the Internet Archive’s automatic transcripts are offered as such. No archive rates radio, so horror, crime, war and westerns are for profiles from 13, racist caricature for adults only with a note saying so, and children’s profiles hear what was made for children. LibriVox’s dramatic readings are in Books.
World Radio
The radio directory comes from Radio Browser, a community-run list of internet radio stations whose data is dedicated to the public domain. Thank you to its volunteers. The streams, names and logos belong to the stations themselves; they play from the stations' own servers, and each card links to the station's site. A station's icon is this site's copy, made the size its card shows it, and only where the site hosting it does not forbid that; otherwise its card shows its name. Stations for adults are for adult profiles only.
Films & TV
Films and cartoons are works in the public domain in the United States or are released under Creative Commons licences, such as the Blender Foundation's open movies (Big Buck Bunny, Sintel, Tears of Steel, Spring). A work licensed by its rights holder plays everywhere; a work in the public domain under US law alone plays where US copyright law governs and where what the site holds shows a place's own term has ended, and its page elsewhere says so (where each plays). Each title's page names its makers, its licence and its source, and captions and subtitles keep the licence of the work they belong to. Live channels are scheduled from the same library.
Live TV from around the world
The channels under Live TV around the world are listed from the iptv-org database (github.com/iptv-org), which its contributors release into the public domain (the Unlicense), read once a day. This site stores and relays none of its streams: each plays in your browser from the server that serves it. A stream plays only where it is shown to be the broadcaster's own: it comes from the broadcaster's own web address (agreeing with Wikidata where the channel's item is linked); the broadcaster's own site plays it, looked at about once a week by a robot that honours robots.txt; its rights holder documents it for anyone to use; or a person traced, as a visitor, who serves it and how it is tied to the broadcaster, citing every page read, and someone else checked the trace again. Each channel's page says which, and links to what shows it. A feed made for one platform, such as Fire TV, Samsung TV Plus or Tubi; a feed a distributor such as Frequency or Wurl serves to the platforms a channel is distributed to; an address carrying someone's access token; a reseller's or restreamer's copy; and any stream not shown to be the broadcaster's own do not play. Those channels stay listed, marked not offered here, and each one's page says why and links to the broadcaster's own site where it has one. Every stream offered is checked from our server, in the United States, about once a day, and a channel plays once one of them has passed; one that plays only in its own country may fail that check and still play there. Left out: channels iptv-org has blocked for a rights holder's copyright notice, closed channels, streams a browser cannot play, streams from sites that restream other people's channels, and channels their owners sell. Channels iptv-org marks as adult are for adult profiles only; a channel of pornography is listed for them without its logo and plays nowhere yet, as the law where it would play asks for a proof of age this site does not have (why). Which profiles see a channel is read from its kind: children's profiles see children's, family and educational channels; from 10, factual ones such as news, documentaries, music and sport; from 13, everything else. A betting channel, named in a table with the evidence for each, is for adults only, labelled, left off the sports rows and what the site suggests, and offered only where its operator's licence is clear (the list). Names, logos and trademarks belong to their owners, who do not endorse this site. A channel's logo is this site's copy, fetched once from the site hosting it and made the size its card shows it, and only where that site allows it: Imgur, which hosts most of the logos the database lists, forbids being used as other sites' "content delivery network", and Pluto TV, ImgBB, YouTube, Amazon, Plex, X, Fandom, DistroTV, Scripps, Sun NXT, Axios, TV Passport, Google Play and Douyin forbid showing their images on other sites or allow it only for your own personal use, so a channel whose logo is there shows its name instead. The same goes for radio stations' icons. To have a channel removed, write to us or to iptv-org.
Films from the world's archives
The films under The world's archives are listed from the Internet Archive (its film, animation, television, educational and ephemeral collections), Wikimedia Commons, the NASA Image and Video Library, the Library of Congress and Open Images (the Netherlands Institute for Sound and Vision), read through their public APIs once a week or month. Only what they say about each film is kept here: every film plays from the archive's own file, and its page names the archive, links to the film there and states its licence or status as the archive does, without claiming more. The archives give no age ratings, so each film is placed from its collections and subjects: children's profiles see cartoons, children's and educational films; from 10, documentaries, science and nature, classic comedies and newsreels; from 13, everything else; sexual content, nudity, gore and atrocity footage only adults. A film its archive files as pornography is listed for adults without its picture and plays nowhere yet: the law where it would play asks for a proof of age this site does not have (why). Nothing is left out for being sexual but a sexual depiction of a child, drawn ones too, which is listed nowhere (how films are placed). Films whose files the archive no longer serves are hidden. To have a film removed, write to us or to the archive.
Comics, magazines and newspapers
Print lists public-domain and openly licensed comics, magazines and historic newspapers from the archives that keep them, read through their public APIs, and shows each page in the site's reader straight from the archive's own image server, at the size the screen needs. Only what the archives say about each work is kept here; nothing is copied or re-hosted. Every work's page names its archive, links to the work there and states its licence or status as the archive does, without claiming more.
The Internet Archive (archive.org): its comics, pulp and magazine collections, with the page images of its IIIF service. Wikimedia Commons (commons.wikimedia.org): public-domain comic strips, each under the licence its file page states. Chronicling America (loc.gov), from the Library of Congress and the National Endowment for the Humanities: historic American newspapers, with the text of every page as the Library read it by machine. Where this server holds a key, Europeana's newspapers (those whose rights statement allows reuse) and Trove's, from the National Library of Australia.
Archives give no age ratings, so each work is placed from its collections, subjects and title, cautiously: children see only what is made for children; horror, crime, war and the pulps are for teens; works with racist caricature, wartime and jungle comics (which commonly carried it), true-crime magazines and "weird menace" pulps are for adults only, with a note saying why; every newspaper is for teens and older. Erotica is for adults only; a pornographic comic is listed for them without its cover and offered nowhere yet (who sees what). Works whose pages the archive no longer serves are hidden. To have a work removed, write to us or to the archive.
Art
The works under Art come from the open-access collections of The Metropolitan Museum of Art (CC0), the Art Institute of Chicago (CC0; its descriptions CC BY 4.0, credited where shown), the Rijksmuseum (Public Domain Mark), the Cleveland Museum of Art (CC0), the National Gallery of Art (CC0), SMK – the National Gallery of Denmark (Public Domain Mark), the J. Paul Getty Museum (CC0; digital images courtesy of Getty's Open Content Program), Wikimedia Commons' featured and quality pictures (each under the free licence its page states, with its author credited), and, where this site is set up for them, the Smithsonian Institution's art museums (CC0) and Europeana (openly licensed works from Europe's museums). Their catalogues are read through their public APIs and data files, politely and not more than every two weeks to a month. What the museums say about each work is kept here, and a copy of its picture at the sizes the pages show where the museum's terms allow one (the Met, Cleveland, the National Gallery of Art, SMK and Wikimedia Commons, fetched once and served from this site); every other picture is shown from the museum's own image server, as the Art Institute of Chicago asks. Cleveland's pictures are also made the size a card shows them at (400 and 800 pixels on the longer side), once each, on the server's CPU, with Apple's ImageIO; a card's picture is then about an eighth of the bytes, which saves more energy in its first view than making it took. Each work's page names its museum, its credit line and its rights statement as the museum states it, and links to the work's page there. The museums do not endorse this site. Museums give no age ratings, so which profiles see a work is read from its title and subject terms: explicit sexual content, nudity, violence and gore only adults; war, the Crucifixion, death and the like from 13; children's profiles only works whose subjects are plainly safe (how works are placed). To have a work removed, write to us or to its museum.
Objects in 3D come from the Smithsonian Institution's Open Access 3D models (CC0, where this site is set up for them) and NASA's 3D Resources (which NASA states are generally not subject to copyright in the United States; only the models NASA names itself the source of). Every model is loaded from its source's own server, and each object's page names its holder, its credit and its licence as the source states it. Human remains and what a museum treats as culturally sensitive are never shown in 3D; their records keep their pages, saying why. NASA does not endorse this site, and its insignia are not shown.
Learn
The courses, lecture series and talks under Learn come from MIT OpenCourseWare (CC BY-NC-SA 4.0), NPTEL of the Indian Institutes of Technology and the Indian Institute of Science (whose pages state "Creative Commons Attribution-ShareAlike / CC BY - NC - SA"), Open Yale Courses (CC BY-NC-SA 3.0), Stanford Engineering Everywhere (CC BY-NC-SA 4.0), Harvard's CS50 (CC BY-NC-SA 4.0), UC Berkeley's courses and other university collections at the Internet Archive (each with the licence the Archive states for it, where it states one), Gresham College (whose terms encourage the use of its lectures for education, not for gain), TED (CC BY-NC-ND 4.0, shown in TED's own player, as its usage policy asks) and CERN (free for educational and informational use, credited to CERN, shown in CERN's own player). Their catalogues are read from their own published pages, data files and APIs, politely and once a month. Every lecture plays from its source's own servers: in this site's player where the source serves the file, otherwise in the source's own embedded player (YouTube's, TED's, CERN's), which loads only when you press Play. Each course's page names its institution and teachers, states its licence as the source does, and links to the course, its syllabus and its readings there. Most of these licences allow only non-commercial use: this site sells, places and inserts no advertising and charges nothing. The sources do not endorse this site. Sources give no age ratings, so which profiles see a course is read from its subjects and titles (how courses are placed). To have a course removed, write to us or to its source.
Picture enhancement
Low-resolution films are sharpened on your own device's graphics chip. Animation uses Anime4K (© 2019–2021 bloc97, MIT License); live action uses AMD FidelityFX Super Resolution 1 (© 2021 Advanced Micro Devices, Inc., MIT License). Turn it off in the player's settings menu.
Games
The games played here are open-source, each under its authors' licence, credited on its page with its source and what was changed, and run from a separate site (invnet.org). The other games listed are facts from Wikidata and links to their makers' own sites; their names and trademarks belong to their owners, and this site does not sell, stream or distribute them.
Legal Disclaimers
Technology Demonstration
Invisible Network is a technology demonstration and proof-of-concept platform. It is not a licensed broadcaster, cable operator, or commercial streaming service. Content available on this platform is provided for demonstration, educational, and research purposes.
No Professional Advice
Nothing on this platform constitutes financial, medical, legal, or professional advice. Content is provided for informational and entertainment purposes only. Consult qualified professionals for specific guidance.
Content Maturity Ratings
Maturity ratings and age classifications displayed on this platform are informational guidelines. They are provided in good faith but are not guaranteed to be comprehensive or authoritative. Parents and guardians should exercise independent judgment regarding content suitability. The family profile and age-gating features are parental tools, not substitutes for parental supervision.
Machine-Made Text Can Be Wrong
Machine transcriptions (the site's subtitles, the archives' transcripts, text read from scans), machine summaries and translations, and the descriptions and chapter names written with AI assistance may contain errors. None of them is a record of what a person said or saw unless its source says it is. Where it matters, check the work itself.
Third-Party Content & Links
This platform may contain links to or content from third-party sources (Project Gutenberg, Internet Archive, etc.). We are not responsible for the accuracy, legality, or content of external sites. Third-party trademarks, service marks, and trade names referenced on this platform are the property of their respective owners.
Signed Releases
Every release of this site's server is built twice, in two clean directories, from the published source, and goes out only if both builds are identical to the byte. It is then signed, with a record of what it was built from and a list of the components inside it, and each of our servers checks the signature before it runs the release. The key that signs them is at /.well-known/release-signing, with the command that checks a release against it (OpenSSH's ssh-keygen). Its fingerprint is SHA256:ThlSv6hyE9UbniITXXAEnwhtEfHbWPf901UoA7OC/Fk.
Data & Privacy
Your preferences, your place in what you watch, hear and read, and everything personalisation learns are kept in your browser (localStorage), and are described on Settings, which can forget them. The servers keep sums per day with nothing about who: what they sent, how fast pages loaded, how many times each work began to play, how often each of home's rows and the header's sections was chosen on the TVs that keep their usual order, what the players noted about energy (the kind of screen and decoder a video began with, eco mode, videos paused because nobody could see them), and the servers' own energy. No personally identifiable information is collected, sold, or shared with third parties. Cookie-based profile selection is used solely for content filtering within the platform.
Regulatory Compliance
This platform is designed with awareness of the following regulatory frameworks:
- EU AI Act (2024/1689) — AI content labeling, transparency obligations
- US FTC Act, Section 5 — Deceptive practices, endorsement guidelines
- FCC Regulations — Broadcasting standards awareness
- China Deep Synthesis Provisions — Synthetic content labeling
- Brazil AI Bill (PL 2338/2023) — AI transparency requirements
- DMCA (17 U.S.C. § 512) — Copyright safe harbor provisions
As a technology demonstration, this platform operates in good faith compliance with applicable laws. If you believe any content on this platform infringes your rights, please contact us.
Limitation of Liability
Invisible Network is provided "as is" without warranty of any kind, express or implied. The platform operators shall not be liable for any direct, indirect, incidental, special, consequential, or punitive damages arising from use of the platform or reliance on any content provided herein.
Where the Site Operates, and Copyright
Invisible Network operates only where the law allows it: what it offers where under US sanctions and the media bans of the European Union, Liechtenstein, Iceland and Ukraine, with every text cited, is on Where the site operates.
To tell the site about a work you believe infringes your copyright, write to its designated agent under 17 U.S.C. 512: Copyright has the agent's details (U.S. Copyright Office registration DMCA-1070530), what a notification must contain, counter notification and the repeat-infringer policy.