AI Video Prompt Guide: How to Write Prompts That Actually Work

Here's the uncomfortable truth about AI video: if your clips look bad, it's usually not the tool. It's the prompt. I've watched people blame Veo, switch to Kling, blame Kling, switch again — when the actual problem was that they typed "a cool car driving" and expected magic.

Prompting is a skill. A learnable one, and honestly not a hard one once someone shows you the structure. This guide covers the formula I use for every AI video prompt, why each part matters, real examples you can copy, and the mistakes that quietly wreck your results.

What makes a good AI video prompt?

A good prompt does one thing: it removes guesswork. Every detail you leave out, the model invents — and it rarely invents what was in your head. "A person walking" could be anyone, anywhere, shot any way. The model picks. You probably won't like its picks.

So the goal isn't to write more words. It's to write the right words: the ones that lock down what's on screen, how it moves, and how it's shot.

The AI video prompt formula

Five parts. Use them in this order and your results jump immediately. This works whether you're doing text-to-video or animating a still with image-to-video.

[Subject] + [Action] + [Setting] + [Camera] + [Lighting & Style]

• Subject — who or what's in frame, described specifically. Not "a woman." A woman in a red wool coat, mid-thirties, dark curly hair.
• Action — what they're doing. It's video. Give it something to animate.
• Setting — where, and what it feels like. "A rain-soaked alley at night, neon signs reflecting in puddles."
• Camera — the shot and the move. This is the one beginners skip, and it's the one that separates a shot from a snapshot. "Slow push-in." "Low tracking shot." "Drone pull-back."
• Lighting & style — golden hour, harsh overhead fluorescents, cinematic, documentary, anamorphic lens flare. Sets the whole mood in a few words.

You don't need every part every time. But the more of them you include, the less the model has to guess.

Bad prompt vs good prompt

Same idea, two very different results.

Bad: "A dog on a beach."

Good: "A golden retriever puppy splashing through shallow waves at sunset, camera tracking alongside at a low angle in slow motion, warm golden-hour light, cinematic, shallow depth of field, water droplets catching the light."

The second one tells the model exactly what to build. The first one asks it to invent a movie from four words. Guess which produces something you'd actually post.

AI video prompt examples you can steal

Adapt these — swap the subject and setting, keep the structure.

Cinematic: "A lone figure in a long coat walking down an empty neon-lit street at night, slow low tracking shot from behind, heavy rain, reflections on wet asphalt, moody cinematic lighting, shallow depth of field."

Product: "A sleek glass perfume bottle rotating slowly on a marble surface, macro close-up, soft studio lighting from the left, clean white background, subtle reflections, premium commercial style."

Nature: "Morning mist drifting through a dense pine forest, slow drone shot rising above the treetops, soft golden sunrise light breaking through the branches, cinematic, serene."

Social / vertical: "A young woman laughing while cooking in a bright modern kitchen, handheld medium shot, natural window light, warm and airy, authentic everyday feel."

Character close-up: "An elderly fisherman with a weathered face looking out to sea, slow push-in on his eyes, overcast diffused light, muted colors, documentary style."

Prompt mistakes that quietly ruin your clips

• Cramming three scenes into one prompt. Eight seconds fits one moment. Want a sequence? Generate separate shots and stitch them into a longer piece.
• Vague adjectives. "Beautiful," "amazing," "cool" — these mean nothing to a model. Describe what you'd actually see.
• Forgetting the camera. No camera direction means the model picks a random angle. Every time.
• Contradicting yourself. "Bright sunny day, dark moody atmosphere" makes the model split the difference badly.
• Expecting one take to be the take. Run the same prompt three or four times. Results vary. Keep the best.

Camera moves worth memorising

You don't need film school, but knowing a handful of shot terms gives you control most people never get. A push-in builds tension. A pull-back reveals context. A tracking shot follows your subject and feels alive. A low angle makes things imposing; a high angle makes them small. Tools like Runway even give you manual camera controls on top of the prompt, but the words alone get you most of the way there.

A few habits that made the biggest difference for me

Write the prompt like you're describing a shot to a cinematographer, not typing into a search box. Change one variable at a time when a result is close — swapping the whole prompt teaches you nothing. And keep the prompts that work in a notes file; you'll reuse them more than you'd think. Prompting well is mostly pattern recognition, and patterns come from reps.

If you're still getting your bearings with the tools themselves, start with our guide on how to make AI videos — it covers the full workflow around the prompting.

Where to learn this properly

Prompting is the highest-leverage skill in AI video, and it's the one thing that carries across every tool — Veo, Kling, Runway, whatever comes next. It's also the thing that's fastest to learn from someone who's already done the reps. Inside the AI Video Generator community on Skool, I share the prompts I actually use, what's working right now, and how people are turning this skill into income. Come join us, and I post walkthroughs on my YouTube channel.

Not in the mood to learn prompting at all? Fair enough — you can just have us create a custom AI video for you.

Frequently asked questions

How do you write a good AI video prompt?

Use the formula: subject, action, setting, camera movement, then lighting and style. Be specific with each part, describe one clear moment, and always include a camera direction — that's the detail most people leave out.

How long should an AI video prompt be?

Usually one to three detailed sentences. Long enough to cover the five parts of the formula, short enough that you're not contradicting yourself. More words isn't automatically better — more specific words are.

Why do my AI videos look bad?

Nine times out of ten it's a vague prompt, no camera direction, or too much crammed into one clip. Tighten the description, pick one moment, and generate a few times before judging the tool.

Do the same prompts work in every AI video tool?

Mostly, yes. The formula carries across Veo, Kling, Runway, and the rest, though each model has quirks and some respond better to certain phrasing. Learn the structure first, then adjust per tool.

Should I generate the same prompt more than once?

Definitely. Models introduce random variation on every run, so the same prompt gives different results. Three to five generations per shot is normal — keep the best one and move on. Google's own Veo documentation makes the same point about iterating on a prompt.

The bottom line

A good AI video prompt isn't longer — it's clearer. Name your subject, give it an action, set the scene, direct the camera, and describe the light. Then generate a few times and pick the winner. Do that consistently and your videos will look better than most people's, regardless of which tool you're using.

Want the prompts that are working right now, plus the workflow around them? Join the AI Video Generator community on Skool — it's where it all comes together.