Assembling a video people watch
Lesson 19 of 22 in the course “AI Video, No Flops”. Free, with animations and a quiz.
The first three seconds
The shots are ready and the budget is intact. Now to assemble them into a video people will finish. The feed’s main rule: viewers decide almost instantly. Both TikTok and Meta, in their ad advice, ask you to grab attention in the first three seconds, or the thumb scrolls on. That means the strongest shot isn’t hidden in the middle; it goes first. No long logo intros and no slow build-ups: hook first, explanations later. That’s exactly why in the last lesson we spent the strong model’s credits on the first shot.
The skeleton of a 15-second ad
A handy starting template for a short ad has four parts. Seconds 0–3 are the hook: Mira mid-jump, motion that stops the scroll. From 3 to 8, the story: the city, the rhythm, a life where this product fits. From 8 to 12, the product up close so it’s remembered. The last 3 seconds are the call to action: what to do next, for example “link in bio”. It’s not a law but a skeleton that makes a first version easy to build. You can give the agent this template right in the brief, and it will sort the shots into parts itself.
One video, three formats
The same video is usually needed in different shapes. Vertical 9:16 (1080 × 1920 pixels) is for TikTok, Reels and Shorts, Mira’s main format. Square 1:1 is for feeds and carousels. Horizontal 16:9 (1920 × 1080) is for YouTube, the website and in-store screens. Decide which formats you need before generating, not after. Why that matters you’ll see on the next slide: turning a finished horizontal video into a vertical one without losses is almost impossible.
You don’t cut a vertical out
Let’s do the maths. A 1080p horizontal frame is 1920 pixels wide and 1080 high. To cut a 9:16 vertical of the same height out of it, you take a strip 1080 × 9 / 16 wide, about 608 pixels. More than two thirds of the frame goes in the bin, and the remaining piece has to be stretched to 1080 wide, so the image goes soft. Worse, the product or Mira’s hand can easily end up past the edge. So shots for vertical are generated in 9:16 from the start: the model builds the composition for a narrow frame and puts the key thing in the centre, where interface buttons won’t cover it.
Three hooks instead of one
Nobody can guess in advance which opening will work, not even pros. So instead of one hook you make several, while the rest of the video stays the same. Here are four proven types. Unexpected motion: a jump, a fall, a sharp turn, and this is where everything you know about body physics pays off. Result first: show the outcome, then how she got there. A question or a line that grabs. And POV, where the camera sees through the hero’s eyes. The agent can easily make three versions of the first shot for pennies in drafts, and you pick which ones to take to final.
Export without surprises
The last technical step is export. For TikTok and Reels a reliable choice is an MP4 file with H.264 video and AAC audio, sized 1080 × 1920. Keep the frame rate the same as the source shots, usually 24 or 30 frames per second: if you change it on export, motion can start to stutter. Let the file name speak for itself: product, hook version, version number and format, for example sneaker_hookA_v2_916.mp4. When you compare a week later which hook worked best, that name will save you a lot of stress.
Checking through the viewer’s eyes
Before posting, Mira watches the video the way viewers will: on a phone. First muted, since many people scroll with sound off, as we discussed in the lesson on sound. Is the story clear from the picture and captions alone? Then on a small screen: do the product and the face read, or does everything get lost? Then paused shot by shot: did an extra finger, mush in the background or a melting logo slip through? And the last item is the AI label and the rights to faces and music. That’s not about editing anymore but about the video not being removed or banned. The final level of the course is about exactly that.
Check yourself
- Analytics show most viewers leave at second two. What should change first?
Correct answer: Put the strongest hook shot first. They leave in the first seconds, so the problem is the hook. The opening must give a reason to watch right away. - Mira made the video in 16:9 and cropped it for TikTok. The sneaker went past the edge and the image is soft. What should she have done?
Correct answer: Decide on formats up front and generate vertical shots in 9:16 from the start. Only about 608 pixels of width are left from the horizontal. A vertical generated from the start keeps both composition and resolution. - The agent made three hook versions: a jump, POV and “result first”. How should she choose?
Correct answer: Release the versions and compare how many people watch past the first seconds. You can’t guess the hook in advance. Viewers give the answer: which version keeps more people. - The video is ready. Which check matters most before posting?
Correct answer: Watch it on a phone with sound off and go through the shots paused. Viewers will see it on a small screen, often muted. Pausing catches defects that fly by at speed.