What you can take, and what you cannot
Ideas, formats and structures are generally free to reuse: a product held up to the camera in the first second, a problem shown before the fix, a close-up demo, a smile with the product at the end. That is how most ad creative is made.
The expression is not free: the footage, the presenter's face and voice, the music, the exact script, and logos and trademarks. Copying those can infringe copyright, trademarks or a person's rights over their likeness. This is general information, not legal advice; ask a lawyer about a specific ad.
| Take from the reference | Replace with your own |
|---|---|
| The order and timing of the shots | The footage |
| The hook's technique | The hook's words |
| Framing and camera moves | The person and their voice |
| Where the product appears, and how big | The product, from your real photos |
| The pace of the edit | The music |
| Whether there is text on screen | The text itself |
Step 1: Measure the cuts
Start from the video, not from a description of it. A scene-change detector finds the cuts: ffmpeg's scene score above 0.25, sampled 10 times a second, works well on ads. Join any shot shorter than half a second to its neighbour, since those are usually flashes rather than shots. On a 40 second test ad, all 11 cuts found this way were real.
An AI model watching an ad is good at saying what happens in each shot and poor at saying exactly when each shot starts. Give it the measured cuts and ask it to describe each one.
Step 2: Write down each shot's job
For every shot, note:
- its purpose: hook, problem, demo, proof, offer or call to action;
- the framing, and how the camera moves;
- the place, and what the person does;
- where the product is, and how much of the frame it fills;
- any text added in the edit;
- what is said.
"Text added in the edit" matters. Without that distinction, a screen recording of an app came back with every button label listed as on-screen text.
Step 3: Study the hook, then write your own
Note when the hook ends, in seconds, the technique it uses (a question, a bold claim, a surprising action, the result first) and why it works for that audience. Then write a new hook with the same technique for your product.
Keep the reference's words out of the rest of the script. When a model was shown the reference's lines for every shot, the remake's lines came back as near copies; shown them for the hook only, as a pattern, it wrote its own.
Step 4: Fit the structure to what you can film
AI video models film clips of a fixed length, about 10 seconds for Gemini Omni, with up to five shots in one clip. Stretch the reference's timeline over whole clips, put the shot edges on whole seconds, and fold any sliver under a second into its neighbour.
A shot cut in two by a clip edge carries on in the next clip. Mark it that way in the script, or its line is said twice.
Step 5: Replace every person, place and line
Write a new presenter, a new place for each part of the ad and new lines in your own words, using only claims your product can support. A different language is fine: a structure that works in English usually works in Hindi or Urdu, with the lines written in that language's own script (why that matters).
Step 6: Keep your product real
Where the reference shows the product alone, use your real product: a still drawn from your product photos and checked against the label, or the photo itself. Where it shows an app, use your real screenshots. A video model left to draw either will invent a label or a screen.
How Hyzgen does it
Hyzgen's Video Cloner follows these steps. Upload an ad of 3 to 90 seconds, and in about half a minute you get its breakdown: format, hook, structure, and every shot's purpose, framing and timing. The upload is deleted once it has been read. The remake is written as a product review around your product, and you can edit every line before anything is filmed. Reading an ad is free on a paid plan; filming costs what a product review costs.