How it works
A pipeline, not a prompt
Specialised agents write the script, record the voice, design or record every scene and check the result before it is rendered. Every stage is saved as it finishes, so an interrupted build carries on from where it stopped.
An explainer, stage by stage
These are the real stages, in the order they run. You watch them tick past in the workspace.
01Reading the brief
One agent settles what the video is actually about before anything is written: the key facts, the terms it will use, and who it is for. If you uploaded documents, they are what it works from.
02Writing the script
The script is a list of scenes: what the voice says in each, the few words that appear on screen, and what is drawn. It is sized to the length you chose, at a calm pace of about 150 spoken words a minute.
03Your review, if you asked for it
With review turned on, the build stops here and shows you the script. Edit it, or type the change you want and get a rewritten script back. Nothing is recorded until you approve it.
04Recording the voice
Each scene’s narration is recorded with ElevenLabs in the voice you chose. The recording decides how long the scene lasts, and its timing is what the captions are laid on.
05Designing each scene
Every scene is designed on its own as animated motion graphics, in your brand colours — with your logo on the first and last — and timed to its own voice. There is no template: two scenes of the same kind in one video are not meant to look alike.
06Checking how it looks
Each scene is opened in a real browser — does it load, does its text stay on screen — and stills of every scene are reviewed for visible defects. A scene that is flagged is redesigned; one that still fails is replaced by a plainer built-in design rather than holding up the video.
07Rendering
Every scene is rendered frame by frame at 1920×1080 and 30 frames per second, so each animation lands exactly where the voice does.
08Assembling the video
The scenes and the voice are joined into one MP4. The captions are burnt in if you asked for them and written as SRT and VTT files either way, and each scene becomes a chapter.
A how-to video, stage by stage
The same voice, captions and assembly — but the picture is a real website, operated in a real browser.
01Exploring the site
A browser opens the address you gave, signs in if you provided details, and looks around the pages the video will need. While it explores it never clicks anything that deletes, removes, pays, sends, publishes or signs out, and it never leaves the site.
02Writing the storyboard
The steps: what to click, what to type, what to point at, and what the voice says during each. Every target comes from what the explorer actually saw on the site, never from what a site like it usually has.
03Your review, if you asked for it
With review turned on, the build stops at the storyboard. Edit the steps, or say what to change and get a rewritten storyboard back. Nothing is recorded until you approve it.
04Recording the voice
Each step’s narration is recorded in the voice you chose, before the browser starts — so every step can be held on screen for exactly as long as it is talked about.
05Recording in the browser
The steps are performed on the real website with a visible cursor and a highlight on what is being explained. If a step cannot find what it should click, the AI looks at the page and picks the matching element; a step that still cannot be done is reported in the chat.
06Cutting and assembling
Long page loads are cut down so nobody watches a spinner, and the voice is moved to match. Captions, optional title cards at the start and end, and one chapter per step — then one MP4, with SRT and VTT alongside.
Sign-in details, handled with care
A how-to video often has to sign in to the tool it shows. This is how the details you give it are treated.
Encrypted, and never shown to the AI
The username, password and company code you enter are encrypted before they are stored. The storyboard only ever contains placeholders; our own code types the real values into the page.
Off camera unless you ask
By default the browser signs in before the recording starts, so the video begins where the work begins. If signing in is what you want to teach, ask for it to be shown — the username will then be visible.
Never shown again
Once saved, the details are never displayed back — not even to you, only whether a video has them. Remove them at any time; they are deleted with the video.
Then keep changing it
A finished video is not a finished conversation.
Ask for a change in the chat — “make scene 3 shorter” — and only the scenes the change affects are made again. A scene whose words did not change keeps its recorded voice, and a scene that did not change at all is not rendered again.
When it is ready, publish it. Publishing freezes a copy: share it by link, or list it publicly on Explore, and keep editing without anyone who is watching seeing a half-made change. They see your next version when you publish again.
Download the MP4 and its SRT and VTT captions at any time, and put them wherever your people watch — any LMS, intranet or video platform that takes a video file.
Make your first video
Describe a topic or name a website. Watch the video get made. Download an MP4 that plays anywhere.
Make a video