A product video shows the product; an ad video asks for something: tap, visit, call, add to cart. That difference changes the process. AI generates the scenes of an ad, but not their order, the first two seconds or the call to action. This article lists seven steps and says where the machine works and where the human does.
1. The one-sentence promise (human)
When the video ends, what will the viewer do and why? “Try our Ramadan menu this week.” “The new-season abaya collection is out.” One sentence. Without it, do not start generating scenes; you get a beautiful, aimless video.
2. The first two seconds (human)
On social media a video has two seconds before it is scrolled past. The first frame must be the product at its strongest, a movement, or a question. An ad that opens with a logo gets scrolled. AI does not make this decision; you do.
3. The scene list (human + machine)
Write three to six scenes that carry the promise: what is visible, what the camera does, how many seconds. Then generate several variants per scene. Short scenes are both where the model is strong and the rhythm of social media. For process and tool criteria see creating product videos with AI.
4. Product accuracy (human)
If the product looks wrong in an ad, the problem is legal, not aesthetic. Colour, cut, logo, stone count — all compared against the reference. Disclosing AI-generated content is also increasingly required; we mark every output we produce as AI-generated in its metadata.
5. Text, voice and captions (machine + human)
Text on the product breaks during generation; text is added in the edit. Voice-over can be AI-generated, but you write the script and check the pronunciation by ear. Captions are mandatory: most social video is watched muted.
6. Format per platform (machine)
- Instagram / TikTok Reels: 9:16, the two-second rule, captions required, short.
- Instagram feed: 4:5 or 1:1; keep the framing centred.
- YouTube and Google video ads: 16:9; the skip button appears at five seconds, so the promise must be said within them.
- Website banner: muted, looped, short; text on the page, not in the video.
Each size gets its own edit from the same scenes; cropping a single video usually pushes the product out of frame.
7. Measure and regenerate (human)
Once live, watch drop-off in the first three seconds, click-through rate and cost per result. Regenerate the weak scene, move the strong one forward. The real advantage of AI production is not the first video but the second and third versions being ready within hours.
When to come to us
For a single campaign you can try it yourself. If you need several campaigns a month, multiple platforms and a consistent brand voice, our social media ads and product video services work together. Write your promise in one sentence in the brief form; we will come back with a scene list.