Most AI video prompts fail for the same reason: they name a topic and nothing else. "How espresso is made" is a topic. The tool has to guess what to show, where to start and when to stop, and a guess gives you the average video on that topic. Below: the parts that change the result, then nine ideas rewritten as full requests.
What makes AI video prompts work?
Five parts. Each one removes a guess.
| Part | The question it answers | Weak | Strong |
|---|---|---|---|
| One subject | What is on screen the whole time? | "Coffee" | "The water in one espresso shot" |
| The hook | Why would anyone stay past second two? | "Did you know?" | "The pump pushes it at about 9 bar" |
| The viewer | Who is this for? | "Everyone" | "People who drink it daily and never saw inside the machine" |
| The path | What happens, in order? | "Explain the process" | "Heated in the boiler, pushed through the coffee, out as two streams" |
| The ending | What is the last image? | "Conclusion" | "The cup filling, the seconds counted" |
One subject is the part people skip. Pick the thing that moves: the water, the key, the air, the light. A subject that moves gives the video a path for free.
The hook is a fact or an image, not a question: see what the first three seconds have to do.
The ending should be something you can see. "End on the door opening" is an ending. "Summarise the key points" is a slide.
How to write AI video prompts for different tools
The five parts hold everywhere. What you add depends on what the tool makes.
- Footage models make a few seconds that look filmed. Add the camera, its movement and the light, and keep to one action per clip.
- Script-to-video and animation tools make a whole short video. Add the length, the voice and the steps in order.
- Editing tools work on your footage. Say what to cut, what to keep and what to add.
What should you leave out of a prompt?
- Numbers you have not checked. A tool that cannot look things up will repeat your number on screen. Leave a blank in capitals, like
[PRICE], and fill it in yourself. - A second topic. "And also mention" is where a tight video becomes a list.
- Style word piles. "Cinematic, epic, stunning, 4K" says less than "slow, close, one colour".
AI video prompt examples: nine ideas, before and after
The facts in these are well established. Where a fact belongs to you, there is a blank.
1. How espresso is made
Before: how espresso is made
After: Show what happens to water in the 25 seconds of one espresso shot. Open on the fact that the pump pushes it at about 9 bar, nine times the pressure of a car tyre. For people who drink it every day and never saw inside the machine. Follow the water: heated in the boiler, pushed through the packed coffee, coming out as two thin streams with crema on top. End on the cup filling and the 25 seconds counted.
2. How a toaster toasts bread
Before: how a toaster toasts bread
After: Cut a toaster open and follow the heat. Open on the wires glowing orange the moment the lever goes down. Show the current heating the wires, the wires radiating heat onto the bread, the surface drying and then browning. End on the latch letting go and the toast jumping up.
3. Why the sea has two high tides a day
Before: why there are two tides
After: Show the Earth with two bulges of water, one facing the Moon and one on the far side. Open on the puzzle: one Moon, two high tides. Follow one beach as the Earth turns: it passes under the first bulge, then low water, then the second bulge on the side away from the Moon. End on the full turn, with both high tides marked.
4. How a pin tumbler lock works
Before: how a lock works
After: Cut a door lock open from the side. Open on the pins sitting across the gap, blocking the cylinder from turning. Follow the key going in: each cut in the key lifts one pin to a different height until all the gaps line up on one line. End on the cylinder turning and the bolt sliding back.
5. Why we always see the same side of the Moon
Before: the far side of the moon
After: Show the Moon going around the Earth with a mark on the face we see. Open on the claim that nobody on Earth has ever seen the other side with their own eyes from here. Use two dials, one for the Moon's spin and one for its orbit. Follow one full orbit, about 27 days, and show both dials finishing together: one spin for every orbit. End on the mark still facing Earth.
6. Why Saturn would float in water
Before: saturn facts
After: Put Saturn in an imaginary tank of water big enough to hold it. Open on the planet bobbing at the surface. Explain why: Saturn is mostly hydrogen and helium, and its average density is lower than water's. Compare it with the Earth, which would sink. End on Saturn floating, rings and all.
7. One heartbeat in slow motion
Before: how the heart works
After: Cut a heart open and slow one beat down to fill the whole video. Open on the four chambers, empty and still. Follow the blood: into the two upper chambers, through the valves into the two lower ones, then pushed out to the lungs and the body. Show each valve closing, which is the sound a doctor listens for. End on the beat starting again.
8. A video for your own product
Before: a reel for my course
After: A narrated reel for [PRODUCT], which helps [WHO] to [RESULT]. Open on the problem they have now: [PROBLEM IN THEIR WORDS]. Show three steps of how it works: [STEP 1], [STEP 2], [STEP 3]. Give the price as [PRICE]. End on the name and where to get it: [LINK].
9. An edit of your own take
Before: edit my video
After: Cut the pauses, the stumbles and the repeated takes out of my recording and keep the cleanest version of each sentence. Keep the first line as the hook. Frame it vertical, add captions word by word, and zoom in on the main point. Put my music file under the voice, quiet enough to hear every word.
Is there an AI video prompt generator or prompt enhancer?
Any chat assistant will expand a short idea if you give it the five parts as a checklist. The risk is that a prompt enhancer fills gaps with invented facts, so read what it wrote before you use it.
Easy Reels has this built in as a button called Improve my idea, on every form. You type a few words and it rewrites them as a full request, with the one thing the video shows, a hook, the steps and the ending. Three things about it:
- It never invents your business. What only you know, like your price or your result, comes back as a blank in capitals for you to fill in, as in example 8.
- Facts are looked up. With Claude Code or Codex, prices, dates and records are researched on the web before the script is written, and a figure it cannot confirm stays out of the video.
- You stay in charge. The improved request is plain text. Keep it, edit any word, or send your own words.
Example 1 above is a real pair from the app.
To start from a finished request, the prompt library has fill-in prompts by kind of video.
Easy Reels is a desktop app for Mac and Windows, $39 once: pricing. It makes animated and narrated reels and edits your own videos. It does not generate realistic footage, so prompts for filmed-looking clips belong in a footage model, not here. Can Claude make videos? explains the difference.
Questions
How long should an AI video prompt be?
Long enough to hold the five parts: usually three to six sentences. Past that, you are often adding a second topic.
Do I need a different prompt for each AI video tool?
The five parts carry over. Add camera and light for footage models, length and voice for script-to-video tools.
Should I put numbers in my prompt?
Only ones you have checked. For anything else, leave a blank in capitals, or use a tool that looks facts up and shows its sources. For formats to start from, see reel ideas.