Where everything comes from

Every picture, film clip, sound and piece of music in our documentaries and story packs is listed here, with its licence, the page we took it from, and proof.

The rules we work by

• Facts: every fact on a story page links to its source. Wikipedia is only used to find sources, never as one.
• Media: only public domain, CC0 or CC BY. Never "non-commercial" (NC), "no derivatives" (ND) or "share-alike" (SA) material, and no clips used under "fair use". Every item is credited on screen and below.
• AI: We use AI openly and say where: scripts are drafted with Claude and checked by hand against the sources on each story page; the narrator is a synthetic voice; translations are machine-made and not yet proofread by native speakers; AI-made illustrations are always stylised and labelled as such on screen.
• Mistakes: if you think something here is wrong, write to hello@echoesof.science. We check it, and fix or remove it.

How to check us

For each item: the source link is the page we took it from. Archived opens a copy of that page saved on the Internet Archive, so you can see what the licence said when we used it, even if the page changes later. The code is the start of the file's SHA-256 fingerprint: hash the file we published and it matches. The full list is downloadable as JSON or CSV.

Right now: 722 items in 29 films and packs: 477 Public domain, 83 CC BY 4.0, 70 CC BY 2.0, 56 CC0, 14 CC BY 3.0, 11 No restrictions, 5 CC BY 3.0 us, 4 CC BY 2.5, 2 CC BY 3.0 de. 342 source pages archived so far (more are added every day). 11 items are under review because their licence label needs a closer look.

What it cost, and its footprint

All of this is paid by echoesof.science. No sponsor and no grant.

24 documentaries, 55.2 hours of rendering in total. That is about as much electricity as one 10-watt LED bulb uses in 954 hours. Electricity is estimated from the PC's power draw, sampled during every render (not a meter at the wall), and priced at our tariff. It counts the final render of each film; test renders, re-renders and the PC idling between jobs are not in it yet. Carbon uses Belgium's average grid mix for 2025, 181.7 g CO₂ per kWh. The energy used by the AI provider's data centres is not published per user, so we can't include it, and we don't guess it. The pictures, film and music themselves cost nothing: they have been sitting in public archives, many of them for decades, free for anyone to use.

Where the material comes from

477 from Wikimedia Commons, 66 from NASA Image and Video Library, 60 from NASA, 43 from Commons, 30 from Wikimedia, 24 from incompetech, 4 from Wellcome Collection, 4 from DVIDS, 3 from The Metropolitan Museum of Art, 3 from Flickr, 2 from Wellcome, 2 from Flickr via Openverse, 1 from arXiv, 1 from Europeana / Wellcome Collection, 1 from Europeana / The British Library, 1 from Historic Oregon Newspapers.

What we make things with

• Claude (Anthropic): researching and drafting the scripts and pages; every fact is then checked against the sources listed on the story page (AI model, used under Anthropic's terms)
• Chatterbox (Resemble AI): the narrator's voice: it is a synthetic voice (MIT)
• Kokoro: backup voice (Apache 2.0)
• Gemma (Google), run on our own PC: translated subtitles and Short titles in 14 languages (Gemma terms of use)
• Blender: 3D scenes (Earth and Moon cut-aways and similar) (GPL)
• Remotion: animated maps, timelines, counters and charts (Remotion licence (free for individuals and small teams))
• d3-geo + Natural Earth: map projections and coastlines (ISC / public domain)
• FFmpeg: cutting, encoding and the film look (LGPL/GPL)
• CLIP (OpenAI): checking that real-footage clips match what the narrator says (MIT)
• Bebas Neue, Newsreader, IBM Plex Mono: fonts (SIL Open Font License)

Everything else is our own work: the scripts, the graphics, the labs, quizzes and games.