The script is cut into beats
The script is divided into scenes and then into short beats — usually a clause of a few words, the length a single picture can carry. For each beat the engine works out what it is about and what should be on screen while it is spoken.
Every beat gets a real timestamp
Your voiceover is transcribed word by word, and the transcription is aligned back onto the beats. Every cut in the finished video lands on your actual delivery — your pauses, your pace — rather than on an estimate of how long a sentence takes to say.
Footage is found, watched and judged
For each beat the engine writes a search, pulls candidate clips from real video libraries, and scores every frame against what the line says. Clips that are diagrams, talking heads, watermarked or too short are rejected. When no clip fits, it falls back to a photograph, screened for watermarks and burned-in text.
On a top-ten list each item gets its own pool of footage, so number seven never plays a picture of number three.
Cards, captions and colour
Where a line states a figure, a year, a quote, a name or a list, the engine can place a design card and fill it with the words from your script. Captions come from the word-level transcription, so they land on your voice. Colour grades and transitions are applied across the cut.
You get an editable timeline
The result is a 1920×1080 timeline (or 9:16 for Shorts, Reels and TikTok) with the footage trimmed to each beat. Open it in the browser editor to swap a clip, retime a beat, drop in your own upload or change the grade — then export. A four-to-five minute video takes twenty to forty minutes, and you can close the tab while it runs.