In this article
- What is an AI video prompt?
- What goes into a good AI video prompt?
- Why does my AI video go wrong?
- How do I build a prompt step by step?
- What camera words should I use?
- What is image-to-video, and why does it help?
- What video prompts work for a UK business?
- What should I check before publishing AI video?
- How to prompt AI video: questions people ask
The short answer
To prompt AI video, describe one shot in six parts: the subject, one action, the setting, the camera move, the light and style, and any sound. Keep each prompt to five or ten seconds of footage. For consistency, start from a still image and describe only the movement.
An AI video prompt is a shot list that has been told to behave. You are not writing a film. You are writing five to ten seconds of one, and the AI is a very keen camera operator with no common sense.
I have watched plenty of people type "make a nice video of my business" and get a horse eating a croissant. That is not the AI being stupid. That is the AI doing exactly what a vague brief deserves. Here is the method that gets you the shot you pictured.
What is an AI video prompt?
An AI video prompt is a short written description that tells a video model what to show: who or what is on screen, what they do, where they are, how the camera moves, and what the light and style look like. Tools such as Google Veo, Runway, Kling and Luma turn that text, or a still image plus text, into a clip of a few seconds.
Two rules sit under everything on this page. One prompt, one shot. And say what the camera does, because if you do not, it will pick something dramatic.
What goes into a good AI video prompt?
Six parts cover nearly every shot. You will not need all six every time, but when a clip goes wrong, one of these is usually missing.
| Part | What to say | Example |
|---|---|---|
| Subject | Who or what, with two or three physical details | A woman in her fifties in a green apron |
| Action | One simple movement, in the present tense | slides a tray of scones onto a counter |
| Setting | Where and when, with one detail you could touch | a village bakery at 6am, flour on the floor |
| Camera | Shot size and one move | medium shot, slow push in |
| Light and style | How it looks, in a phrase | warm window light, 35mm film grain |
| Sound | If the tool makes audio, say what you hear | oven fan, radio murmuring, no music |
A woman in her fifties in a green apron slides a tray of scones onto a wooden counter in a village bakery at 6am. Flour on the floor, steam rising. Medium shot, slow push in. Warm window light, 35mm film grain. Sound of an oven fan and a radio murmuring in the background, no music.
Why it works: one subject, one action, one camera move. Nothing is competing.
Why does my AI video go wrong?
Almost always because the prompt tried to be a film. Ask for a character, a chase, a costume change and a sunset in one go and the model blends them into something that looks like a fever dream. Keep to one shot of five to ten seconds. Stitch shots together afterwards in an ordinary editor.
| What went wrong | Why | Say this instead |
|---|---|---|
| Wobbly, morphing hands and faces | Too much movement at once | Reduce to one slow action. Add "locked-off camera" |
| Camera swoops about for no reason | No camera instruction given | Name one move: "slow dolly in" or "static tripod shot" |
| Looks like a stock advert | Style left to chance | Add a film stock, lens or a lighting condition |
| Same person looks different in each clip | Each prompt starts from scratch | Use a reference image. See image-to-video below |
| Text on signs is gibberish | Models still struggle with lettering | Leave signs blank and add real text in an editor |
How do I build a prompt step by step?
Do not start with the clever version. Start plain and add one thing per attempt, so you know what changed the result. Here is a shot of a farm shop going from forgettable to usable in three rounds.
A farm shop with vegetables.
A farmer in a flat cap arranges crates of Cotswold carrots on a wooden stall outside a farm shop on a frosty morning.
A farmer in a flat cap arranges crates of Cotswold carrots on a wooden stall outside a farm shop on a frosty morning. Low sun behind him, breath visible in the cold. Medium shot, slow push in, shallow depth of field. Natural sound only: crates scraping, a distant tractor.
Round 1 gets you something. Round 2 gets you something yours. Round 3 gets you something you can use. Three attempts is a normal day, not a failure.
What camera words should I use?
You do not need film school. A short list of phrases covers most needs, and the models understand them well.
| You want | Write |
|---|---|
| Calm and steady | static tripod shot, locked-off |
| To draw the viewer in | slow push in, slow dolly in |
| To reveal a place | wide establishing shot, slow pan left |
| Intimacy | close-up, shallow depth of field |
| Following someone | tracking shot behind the subject, steady |
| A sense of scale | low angle, wide lens |
| The overhead view | top-down shot, slow rotate |
What is image-to-video, and why does it help?
Most tools let you upload a still image and animate it. This is the single best trick for consistency. Generate or photograph the frame you want once, then ask the model to move it. Your character, shop front or product stays the same because it is the same picture.
- Make the stillUse a photo you own, or generate a still image until it looks right.
- Describe the movement onlyDo not re-describe the picture. Say what changes: "Steam rises, she turns to smile, slow push in."
- Keep it shortFive seconds of good beats ten seconds of drift.
- Reuse the stillFor the next shot, start from the same image or a close variant.
Animate this photograph of a ceramic mug on a pine table. Steam curls from the mug. A hand enters from the right and lifts it slightly. Static camera, soft morning light, no change to the mug's design or logo.
What video prompts work for a UK business?
Use AI video for what it is good at: short atmospheric clips, backgrounds, social teasers and concept shots. Do not use it for anything where the real thing matters, such as a property you are selling or a product people will buy. For those, film the real thing.
Aerial drone shot rising slowly over a Cotswold village of honey-coloured stone cottages at golden hour, church tower in the centre, smoke from one chimney. Smooth, cinematic, no people, no text.
Close-up of a pint being pulled in a low-beamed country pub, warm lamplight, condensation on the glass. Cut feeling, shallow depth of field, murmur of conversation and a log fire crackling.
Slow tracking shot along a tidy desk with a laptop, a notepad and a cup of tea, rain on the window behind. Soft grey light, minimal, no faces, 8 seconds, loops cleanly.
What should I check before publishing AI video?
- Licence terms. Check the tool's commercial-use terms on the plan you are on. They differ between tiers.
- Real people. Do not generate or animate someone's likeness without their permission, and do not make a real person say things they did not.
- Disclosure. Platforms increasingly ask for AI-generated content to be labelled. Do it. It is cheap insurance.
- Brand and trademarks. Models will happily draw a famous logo. Do not publish it.
- Your own data. Do not upload photos of customers or staff to a tool you have not vetted. Our guide on keeping information safe using AI tools covers the basics.
How to prompt AI video: questions people ask
What is the best way to write an AI video prompt?
Describe a single shot in six parts: subject, action, setting, camera, light and style, and sound. Keep it to one action and one camera move. Add detail one element at a time so you can see what changes the result.
How long should an AI video prompt be?
Between 40 and 80 words suits most tools. Shorter prompts leave too much to chance. Longer ones tend to introduce contradictions the model cannot resolve.
Why do the people in my AI videos look different each time?
Each prompt is a fresh roll of the dice. Start from the same reference image in an image-to-video workflow, or use a tool's character reference feature if it has one.
Can I use AI video for my business?
Yes, for mood pieces, backgrounds, social teasers and concepts. Check commercial licence terms on your plan, label AI content where platforms ask, and film the real thing when authenticity matters.
Which AI video tool should I use?
It changes often. Google Veo, Runway, Kling and Luma are among the established options. Try the free tier of two, run the same prompt through both, and pick the one whose look suits your work.
Do the same prompting skills work for text and video?
Mostly yes: be specific, show what you want, set limits. The difference is that video prompts need camera and movement language. Our top 10 prompt tips cover the shared habits.
Keep reading
- AI prompt engineering: top 10 tipsThe habits that make every AI tool behave better.
- Best AI tools for content creationWhere video fits in a content workflow.
- AI tools for social media marketingPutting clips to work on the channels that matter.
- AI marketing without being spammyKeeping the human in the work.
Want your marketing team using AI without the croissant horse?
We train UK marketing teams to brief AI properly across text, images and video, and to keep the brand voice intact while they do it.
- AI x Marketing Consultancy Hands-on help with your real campaigns.
- AI x Brand Guidelines Teach the AI your house style.
- Programmes Every course in one place.
Harry Lang
Harry Lang founded The Oxford AI School in Witney, Oxfordshire. He trains UK businesses and teams to use AI well, writes the Marketing Unfiltered newsletter, and is the author of Learn To Fly With AI.