Skip to main content

​
Overview

An agent can make a short animated explainer video for you about almost any topic: how something works, how it came to be, a process, a place, an idea, your own product, or “how our lineage works”. It writes the story, designs a drawing for every line, draws it in one of two styles, records an AI voice-over, adds a music bed and renders MP4 files you can download. You approve the story, the script and the key frames before anything is rendered, and you get a quality report with the files. The agent never publishes a video. It hands you the files and a draft caption; posting is your decision.

​
Ask for a video

Ask in chat, in your own words. For example:
Make me a whiteboard video about how a sailboat sails upwind.
Make an explainer about how the Channel Tunnel was conceived, built and operates.
Make a 30-second social short, in the orange orb style, about catching bad data before it reaches a dashboard.
The agent uses its explainer-video skill and works through four steps, stopping at each one for your approval:
  1. The message. One sentence that is the point of the film, and 3 to 5 beats that carry it. Agreeing this first keeps the story from drifting.
  2. The script, the drawings and the stills. One narration line per beat (60 seconds or less unless you ask otherwise), and for each line a plan of what the viewer sees. Facts come from your own material and reliable sources the agent can reach (a web search, when one is set up), kept modest, with no invented numbers or dates. In a film about Genesis, every product claim is checked against the Genesis documentation; in a film about your data, every number comes from a real query or is marked as illustrative. The agent records the voice-over and shows you a still of every beat. The film also plays in its app: open the app to play it, scrub through it and switch between 16:9 and 9:16.
  3. The render. The agent renders the films in the background and checks them automatically (see What the quality check covers).
  4. Delivery. You get the MP4 files, a draft caption that says the voice is AI-generated, and the quality report, including what it could not check.
Ask for changes at any step: a different line, a different scene, a shorter film, the other style. The agent edits the storyboard and redoes only the steps the change touches.

​
What you get

Every film has the voice-over, burned-in captions (most social video plays muted), and a generated music bed under the voice, ducked while the voice speaks and mixed to about -14 LUFS. The music has no percussion. In a film about Genesis no third-party logos appear on screen; in a film about your own organization the agent follows the brand you give it.

​
Styles

Say which you want (“whiteboard”, “the orb style”, “with little stick figures”), or let the agent suggest one: whiteboard for teaching and step-by-step explanations, orb for a punchy social short. The style decides how things are drawn and the storyboard what is drawn, so the same film can be switched to the other style.

​
How the drawings are made

For every beat the agent decides what you need to see to understand the line, then draws it the simplest way that works:
  • A preset scene of the style (a row of boxes, a list of steps, a checklist, a closing card), filled with the film’s own words.
  • A composed drawing built from general elements, all drawn by the style’s own pen:
The template ships two films made this way as examples: “how a sailboat sails upwind” and “how the Channel Tunnel was conceived, built and operates”, each in both styles.

​
Pictures and transparency

When the pen cannot draw something convincingly (an animal, a particular machine, a building), the agent can make a picture with AI image generation. It describes only the subject; the style’s look is added automatically, and pictures never contain words. The picture comes back with a transparent background, so it sits on the board like a drawn object and can fly in, move along a path or be sketched in by the hand:
  • on GPT-Image 1 models the background is transparent as drawn;
  • on other image models the picture is drawn on a plain background, which is then removed from the edges inwards. Background enclosed by the subject (the hole in a ring) can stay, so solid subjects work best.
The agent looks at every picture in the stills and redraws or replaces one that came out wrong. Each picture is a paid image request, so a film usually has none to four. You can also give the agent your own images (a logo, a product photo) to use in the film.

​
Longer films

Films are 60 seconds or less by default. Ask for a longer one, up to five minutes, and the agent splits it into chapters, each opened by a title card.

​
Prerequisites

The agent checks these itself before it starts and tells you plainly what is missing. Without Chromium a video cannot be rendered; the agent can still write the script and storyboard, which you can watch in the app’s preview.

​
Air-gapped servers

Making a video needs no internet access when:
  • the voice is the offline Kokoro voice, with its voice bundle (about 350 MB) copied onto the server ahead of time and text_to_speech.local_allow_download turned off. The steps are in GenesisApps Tools;
  • Chromium is part of the server image. Genesis Desktop downloads its browser on first use, so install it in Settings → Advanced → Browser tools while a connection is available.
The fonts, scenes, icons and music are part of the app template, so nothing else is fetched while a video is drawn or rendered. AI pictures need the image provider; on an air-gapped server without one, the agent draws with shapes, paths and icons.

​
What the quality check covers

After rendering, the agent runs an automatic check of the files and gives you its report. A person must watch and listen to every final cut before it is shared. The voice is AI-generated: say so when you post it (the draft caption already does).

​
For developers

The workflow is the explainer-video agent skill. A video is a GenesisApp scaffolded from the explainer_video template: its storyboard.json lists beats, each a preset scene of a style pack or a compose beat of generic elements, with its narration line, described in the app’s STORYBOARD_SCHEMA.md. The template ships ready-to-render sample storyboards in samples/: sailboat.json and channel-tunnel.json (compose beats only, with -orb copies in the other style), and whiteboard.json, lineage.json and orb.json (presets). The explainer_video_action tool narrates, makes and imports pictures, searches the icon set, takes stills, renders and checks the films; see GenesisApps Tools.