AI Flow: Video

Open source under the MIT License. The full Rust source is on this page; build it with ./build.sh.

Turn a board build in Adom Hydrogen into a narrated walkthrough video: on the project's wiki page, and on YouTube when you want it there. Ask for it when you start a build, leave it running, and the video is ready to review when you are back.

adom-wiki pkg install adom/adom-aiflow-video
adom-aiflow-video --help

Or just say it to your AI: "use aiflow-video to make a video of my board."

An example, made with it

John Lauer narrating his Ghost Hand board (project page, on YouTube). The film was cut by the AI, then John recorded the narration on the narration page below: his face in the corner, circles drawn live while he talked, sound from a wireless mic lined up afterwards, captions written from his own words.

Narrate it yourself

The AI voiceover is fine for a draft. A film people want to watch has a person in it. narrate opens a page where you record your own voice over the film, and the AI puts it all together afterwards.

• Your face in the corner, if you want it. Turn the camera on and you see yourself on the film, exactly where you will appear. Drag it anywhere, drag a corner to resize it, and pick a shape: landscape, portrait, square or circle. Zoom and crop sliders keep your face and drop the room behind you. Turn it off for voice only. Nobody has to show their face.

• Draw while you talk. Drag on the film to circle or point at something. The stroke is teal, fades out after two seconds (Space wipes it sooner), and is drawn into the final film at the moment you drew it.

• Two microphones, best one wins. The page records your computer's microphone. If you also hold a wireless mic that records on the device (or a phone), say so on the page: afterwards plug it in and ask your AI to fetch the clip. It lines the better recording up with the page's take and uses it, so the film gets the best sound you made.
• Microphone check. Calibrate reads a line from your own film for about 15 seconds, sets the level for your voice and tells you what it changed. A clock, a progress bar with chapters and a wrap-up warning keep you on pace.
• The film cannot stall on you. Before a take the page loads the whole film into memory and test-plays it out of sight. If your computer stalls on the film as it is, the page switches to a copy in another format that plays smoothly. Every take also logs film time against recording time, so a take is never lost to a hiccup.
• Your AI keeps you posted. A "From your AI" box on the page says what the AI is doing with your take: fetching your recorder's sound, lining it up, putting your voice, picture and drawings on the film, writing captions from what you said, and where to watch the result.
• Review before anything leaves your computer. Play a take back with the film, record again, or start over. Nothing is published until you say so.

YouTube

• Title and thumbnail first, together with you. Before you record, the AI brings a handful of titles and two or three thumbnails, and the narration page shows each pair the way YouTube will. You pick, or ask for other ideas. What the thumbnail promises is what your opening lines should deliver, so this comes before the voice.

• The description. What the film shows, chapters, and a link back to the project's wiki page.

• The upload. youtube check says which channel each of your adom-google accounts uploads to, and how to set one up if you have none. youtube upload always sends the film as unlisted, so only people with the link see it and going public is your call. It then opens the video's YouTube Studio page for you and uploads the film's own captions, because YouTube's automatic ones misspell names. When adom-google cannot reach your channel, the AI uploads through YouTube Studio in a browser it drives.

• You in the thumbnail, pointing at what you built. A pose shot on the same page: your camera with the product's picture on it and a big "POINT AT THIS" marker, a countdown and three stills. The AI cuts you out of your background and puts you in the thumbnail pointing at the product. The helpers for that ship with the app (thumbkit).

The 3D walkthrough

A dynamic skill

AI Flow: Video is a dynamic skill, like AI Flow itself: a small Rust binary that holds the AI's hand. The AI does about 97 % of the work; the binary is about 3 %, the guide and the checker. It suggests the next step, records clips, ships the routine parts as ready-made code and gives feedback as gentle suggestions. It is never a gate: every command completes, and the AI and the human decide.

What the film shows

• The starting prompt, typed into Hydrogen.
• The build's highlights, a few seconds each: parts being found, placement, routing, checks, silkscreen.
• A 3D walkthrough in the Adom 3D viewer: the enclosure part by part, then the board's major parts on the top and the underside, each one highlighted, labelled and described.
• How it was made: which Adom tools did each step, with a few "ask your AI to ..." lines so the viewer can do the same.
• The parts story and the cost: how many unique parts, where they came from, what one board costs and what a hundred cost each.
• Where the files are: the project's wiki page.

What it helps the AI do

Commands

init [--release <board-release.json>] [--target <machine>]    next    steps    ledger
disk --clips N --seconds S      clip start|stop      mark <step> --note "..."
script <script.json>            scan <files...>      track <path>      clean [--yes]
director --out <file.js>        guide                narrate --video <draft.mp4>
brand [--video <final.mp4> --out <branded.mp4>]
youtube check | title "..." | thumbnail <file> | upload --file f.mp4 --title T --description-file d.txt --thumbnail t.jpg | open <link> | captions <link> --file f.srt
narrate pitch --titles "A" "B" --thumbs a.jpg b.jpg      narrate pose --thumbs product.jpg      thumbkit --out thumbkit/

Every command answers OK: with Hint: lines. The skill is in skills/adom-aiflow-video/SKILL.md; the walkthrough guide is in docs/walkthrough.md.

Branding

Your film is yours: it carries no Adom logo. The Adom logo sting is only for official Adom videos made by Adom employees, and brand tells the AI which case applies before it offers one.

Status

0.7.1. It grew out of the Ghost Hand films (john/ghost-hand). Working today: the flow and its hints, clip recording, the disk estimate, script feedback, the text privacy scan, file tracking and cleanup, the walkthrough director, the narration page (voice, camera picture, drawing, playback check, AI notes), and the YouTube channel check, title and thumbnail feedback and upload. Still done by the AI by hand, with the skill as the guide: the cut itself, putting the voice, picture and drawings on the film, lining up a second recorder, composing thumbnails, and the frame-by-frame privacy check.