Sponsored by
Can you still spot AI video in your feed?
It's no secret that I lean on AI across a lot of my work. If you've watched my videos you've seen it in there, and a fair few of you have dropped into the comments asking how I do it.
There are a lot of moving parts, more than one edition can hold, but I'm going to make a start.
This is one piece of it: how those scroll-stopping AI hooks get built, from the first idea to the finished clip.
But first, some catchup on AI this week.
Partner with us
NEWS NEWS NEWS
AI CHEATSHEET 101
You're probably using Claude Code like a chat box.
The cheatsheet puts all 26 slash commands in one place, including the power moves most people never turn on, like /goal to run it to a finish line unattended and /remote-control to steer it from your phone.
3-Second Test
Two things decide whether a reel travels, and the effects are the smaller of them.
The first is the subject.
It has to be something people are already curious about that nobody explains plainly, because that gap makes a thumb stop.
The second is the first 3 seconds, which carry the feeling of that subject before the viewer has chosen to stay.
Most people then make the same mistake.
They open a video generator, type "man on fire", and wait for something cinematic.
What comes back is a melted, gooey mess, because the tool is trying to imagine the picture and the movement at once and manages neither.
The fix is to split the job.
You lock the still image first, then animate it. Get a single frame right and the rest is easy.
The Method
Five steps, whatever the idea.
1. Pick a subject with a curiosity gap, and a first frame that carries the feeling.
For the run-through here, a calm tech founder sits dead still while fire bursts around him. The contrast is the hook.
2. Lock the still before you touch a video tool. A good frame is the easy part to animate later.
3. Let Claude write the prompt.
Describe the shot in plain English and ask for a detailed image prompt. It hands back the lens, the framing, and the way firelight should sit on the face, wording you would never write yourself, and that is why the picture lands.
4. Run the same prompt through two image tools and keep the winner.
Here that's Nano Banana Pro against GPT Image 2, vertical, 4K. GPT Image 2 comes back too sharp, with skin that tips into plastic.
Nano Banana Pro holds real texture and lets the firelight wrap the face. Nano takes it.
5. Animate the same way. Back to Claude for a motion prompt, drop the winning still in as the first frame, then test Seedance 2.0 against Veo 3.1.
Veo starts the founder blinking and breaks the illusion. Seedance keeps the face still while the fire sweeps in, so Seedance wins.
That's one hook. Fire reads as danger so the thumb stops, and then the calm face in the middle of it doesn't add up. That flicker of "wait, what is this" buys you the next few seconds.
Two More, And The Part Worth Stealing
The second hook drops the danger and still grips.
A businessman types calmly at his desk while the whole office sits underwater, fish drifting past, papers floating up through the light.
On the still, Nano Banana Pro again holds the heavy weight of the water while GPT Image 2 goes painted and dreamy.
On the motion pass it flips: Veo handles the floating physics better than Seedance, so Veo takes this one.
No single tool wins every shot, so you test both every time and keep whatever nails that frame.
The clip grabs you for the same reason the fire did. A man working calmly underwater is something your brain has never filed, so it stalls for a beat, and the stall is the hook.
The third hook adds the move worth stealing.
A glowing green algae filter built into a luxury car's exhaust in a dark garage, with one extra line in the prompt: a real person crouched beside it, pointing at it.
GPT Image 2 dropped the filter entirely. Nano Banana Pro nailed it, down to the crouched figure.
That person is doing the work. Put a real human touching or pointing at something impossible, and the brain stops doubting it.
It's the human anchor, and it's the cheapest credibility you can buy in a fake shot.
My Take
The question I get most in my comments is which tool I use.
It's the wrong question.
I watched the best tool change three times in one video. Nano Banana Pro won the stills, then Veo took the underwater shot, and Seedance took the other two.
The winner kept moving.
A tool that wins one shot and loses the next was never the thing doing the work.
The idea was.
A man sitting calm inside a fire stops your thumb because someone pictured it first, and any decent model could have drawn it.
So stop shopping for the perfect generator. Go find an idea worth the effort. It never gets cheaper, and no update is going to hand it to you.
Until next time,
Vaibhav 🤝🏻
If you read till here, you might find this interesting
#AD 1
Moda is the AI design agent with taste
Moda is an AI design product where you prompt what you need, get a complete on-brand design, and edit every element on a full canvas.
Our viral launch hit 4.4M views in days, tens of thousands signed up, and executives at major finance and tech companies now use it.
#AD 2
The best voice models, now across all channels
Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.
ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.
Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.
The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.
