AI Video for TikTok, Reels, and YouTube Shorts: Where Watch Time Actually Gets Decided

Whether a short video gets watched or scrolled past gets decided before the viewer even registers they made a choice. One second, sometimes less. Whatever you packed into your intro, your logo, your nice little title card, it's already lost by then.

That's the whole point of this post: short video lives or dies in that first instant. I'm Alex, I handle content at Flami. I generate vertical clips constantly, for myself and for testing our templates, so I've got a decent pile of results to go on. Sharing it here, and some of it isn't what you'd expect.

Vertical isn't "the same video, just narrower"

The mistake I see most often: someone takes a finished horizontal video, crops the sides, and drops it into Reels. It looks like a compromise because it is one.

A 9:16 frame is its own composition. The product sits centered, large, filling nearly the whole height. No bars at top or bottom, no letterboxing. When you generate a clip for social, set 9:16 from the start instead of reworking it after. The major models all handle it natively: Veo 3.1, Kling 3.0, Runway, and Seedance 2.0 all output vertical out of the box.

Watch out for the safe zone too. Captions, the like button, the creator's name, the platform's own UI, all of it sits along the bottom edge. Anything important there gets covered. Keep the action in the center of the frame.

How long should it actually be

Different platforms pull in different directions here, and there's no single right answer. I'll just tell you what I've seen.

TikTok and Reels reward very short clips. A 9 to 15 second video that gets watched all the way through and looped reads as a win to the algorithm. Longer clips get finished less often, and completion rate is the real currency in short-form video.

YouTube Shorts allows for noticeably longer clips than TikTok and Reels, so check YouTube's current guidelines before locking in a target duration. But push it too far and it still backfires.

My personal range: 9 to 15 seconds for TikTok and Reels, and I let each YouTube Shorts idea dictate how much longer it needs to run. If your idea won't fit, that usually means you're cramming two stories into one clip. Don't. Split them.

One honest caveat: this shifts. Platforms tune their algorithms constantly, and what worked last quarter can underperform six months from now. Watch your own analytics, not just advice like mine.

A structure that holds attention to the end

I split a short video into three parts, and they're deliberately uneven. Equal thirds is how an AI script would structure it. Real attention doesn't work that way.

The hook eats most of the attention budget, so it gets the most thought. This is the very opening, the first instant, and something has to happen there that makes a thumb pause mid-scroll. A sharp motion. An unexpected angle. A bold promise of the payoff, or a blunt question. For product videos, "result first" works best for me: show the payoff, then show how you got there. Just don't open with "hey, today I'm going to show you," that kills the clip by second two. I've burned myself on this more than once.

The middle section runs shorter. The product in action: it works, it goes on, it gets applied, it gets unboxed. Fast, no lingering shots, or people swipe away.

The ending is about one thing only: don't waste the last frame. Skip "link in bio," the button is already right there under the video. Leave a frame worth rewatching. A second view counts in your favor too.

Turning that structure into a prompt

To make this concrete, here's how I write that structure into an actual prompt. Timed out by the second, hook and line included, built for a 9-second TikTok clip. A different product than the blender from the other prompt guide, a power bank:

"Vertical UGC-style video, 9 seconds, shot on a phone, a guy walking down the street.
0-2 sec, hook: he holds his nearly dead phone up to the camera, 3% on the screen, says flatly: 'Dead again, every single day.'
2-5 sec: he pulls a thin power bank from his pocket, snaps it onto the back of the phone via magnet, it sticks, a light comes on, the battery percentage starts climbing.
5-9 sec: he slides the phone with the power bank still attached into his jacket pocket, keeps walking, tosses off to camera: 'Done. No more hunting for outlets.' Final shot: the power bank holding steady on the phone.
Style: raw UGC feel, 9:16 vertical, daytime street lighting, handheld camera, no text overlays."

A relatable pain point in the first two seconds, that pain visibly solved in the middle, a strong closing frame. For YouTube Shorts I'd stretch this same script to roughly 15 seconds and add one more beat.

If you want more on building prompts like this, I've got a full breakdown of the process that goes deeper than this one example.

Sound half your viewers won't even hear

Short video has a built-in contradiction: sound matters, and a huge portion of people have it off. Scrolling in bed, on a commute, in a waiting room, muted is the default for a lot of viewers.

So the rule is simple. Put anything essential on screen as text. If the whole point of the video lives in a voiceover and someone's watching muted, the point is gone. I've trained myself to put the main line as a large on-screen caption no matter what.

That said, don't skip audio entirely, it pulls real engagement from everyone who does have sound on. Veo 3.1 helps here with native audio and sound effects generated right alongside the video, no separate dubbing pass needed. More on Veo here. If you just need background music, stick to licensed tracks or AI-generated audio. Dropping in a popular copyrighted song is a quick way to get a clip muted or taken down.

One video for all three platforms, or three separate cuts

The tempting shortcut: make one vertical video and post it everywhere. Sometimes that's genuinely fine, a solid 9:16 base works across all three.

But if you want the best result on each, a few small adjustments help. TikTok and Reels forgive, even reward, a rawer, more UGC-style look, while YouTube Shorts tends to tolerate a cleaner, more polished cut. Adjust length per platform while you're at it.

You don't need to reshoot from scratch for this. Pull two versions out of one generation, a different length and a slightly different pace. Runway and its Aleph model make this easy, you can take a finished clip and spin out variations by editing details with text, no full regeneration needed.

Building this without an editor or a production team

If you've got one or two products and a clear idea, the simplest path is generating the clip directly for the platform you're targeting on Flami: pick a model, set the format to 9:16, choose your length, write a prompt built for a vertical scene. I've got a separate guide on prompt structure, and there's a set of ready templates by product type in this prompt collection.

If you've got a bigger catalog and need a month's worth of content fast, the "Video from Product" mode builds a batch of vertical clips straight from your product listings using templates. Upload the links, pick your social format, get a stack of clips back. From there it's just picking covers and captions.

The one thing I keep coming back to: don't polish the visuals at the expense of that first second. A rougher video with a strong hook beats a flawless one nobody finished watching.

Build your vertical video → flami.pro

FAQ

What video format do I need for TikTok, Reels, and YouTube Shorts? Vertical 9:16. Every major AI video model (Veo, Kling, Runway, Seedance) generates it natively. Don't take a horizontal clip and crop it, build it vertical from the start.

What's the ideal length for a short-form video? 9 to 15 seconds for TikTok and Reels, and somewhat longer for YouTube Shorts depending on the idea. Higher completion rate means a better algorithm boost, so aim for clips people actually watch to the end.

Does a short video need sound? It helps, it pulls stronger engagement from viewers who have audio on. But most people watch muted, so put the key message on screen as text regardless. Veo 3.1 generates native audio alongside the clip.

Can I post the same video on all three platforms? A basic vertical cut, yes. For better results, make two versions: a rawer, shorter one for TikTok and Reels, a slightly more polished, longer one for YouTube Shorts.

What music is safe to use? Stick to licensed tracks or AI-generated audio. Using a popular copyrighted song is a fast way to get your clip muted or taken down.

Sources

1. Flami: how to write a prompt for AI video
2. Flami: Veo 3.1 overview
3. Flami: Runway Aleph overview