How to Write Better Prompts for AI Video Tools (With Ready-to-Use Examples)

You've typed "dog running on the beach" into an AI video tool, hit generate, and watched the result with a sinking feeling. The dog sort of looks like a dog. The beach sort of looks like a beach. The legs bend the wrong way. There's no sound. You close the tab and tell yourself you'll come back later. Sound familiar? The model isn't the problem. The prompt is — and the fix is the same across the major AI video tools available today.

What these tools actually do

An AI video generator takes your text — called a prompt (the instruction you type to the AI) — and produces a short video clip. Most current models in this category do one of two things: they generate video straight from text, or they take a reference image and animate it.

Some recent releases also generate native audio — meaning the tool builds sound, voice, or music straight into the clip rather than leaving silent footage for you to score later. Others still produce silent video and expect you to add audio in a separate editor. Check your tool's documentation to know which kind you have.

That detail matters for your prompt: if your tool can generate audio, the "audio" slot below becomes a real lever. If not, you can simply skip it.

The six-slot formula

A useful prompt for AI video tends to cover six slots. You don't need fancy language — plain English works fine. Fill in each slot and you'll notice the difference immediately.

1. Subject — who or what is in the shot ("a golden retriever")
2. Action — what they're doing ("running through shallow water")
3. Setting — where it happens ("a quiet beach at sunrise")
4. Camera — how it's filmed ("low angle, close to the water, tracking the dog from the side")
5. Motion — how things move ("steady, smooth, slow motion at 50% speed")
6. Audio — what we hear ("gentle waves, soft wind, no music") — include only if your tool supports it

Compare the original "dog running on the beach" with the filled-in version. Same idea, much more to work with.

Step 1 — Open your AI video tool and start a new project

Pick the AI video tool you have access to (common options include Runway, Pika, Kling, Luma, and several others built into platforms like Adobe Firefly). Open it in your browser or app, and look for a button or field labeled something like "Generate," "Create," or "New video."

Tap that, and you'll usually land on a screen with a text box for your prompt. Some tools also offer an image upload field next to it.

You'll know it worked when you see an empty text box (and possibly an upload area) waiting for your prompt.

Step 2 — Fill in the first three slots

Write one sentence that covers the subject, action, and setting. These three give the model the basics: who, doing what, and where.

💬 Example: "A barista in a small coffee shop, pouring latte art into a cup, warm morning light coming through the window."

Don't worry about the camera or motion yet — get the scene locked in first. If the basic composition comes back wrong, no amount of camera detail will save it.

You'll know it worked when you can read your prompt back and someone else could sketch the scene from your words alone.

Step 3 — Add the camera and motion slots

Once the basics feel right, add how the shot is filmed and how things move. This is where most beginner clips look flat — the prompt described what was in the frame but said nothing about how the camera behaved.

💬 Example: "Close-up on the hands and cup, slow smooth motion."

Pick one camera style per clip. Mixing "tracking shot" with "static wide" in the same prompt usually confuses the model.

You'll know it worked when your clip has a sense of direction or pace, not just a held frame.

Step 4 — Add audio if your tool supports it

Some current tools generate audio natively; many still don't. If yours does, describe what should be heard. If it doesn't, skip this slot — the model will either ignore it or throw an error.

💬 Example: "Gentle cafe background sounds and a soft espresso machine hiss."

If your tool produces silent video, you can always add sound later in a free editor like CapCut, DaVinci Resolve, or even the audio tools inside Canva.

You'll know it worked when you watch the clip with sound on and the audio matches the scene.

Step 5 — Generate, watch, and change one slot

Hit generate, watch the result, and don't rewrite the whole prompt if it isn't right. The fastest way to improve is to change one slot at a time between tries. Swap the camera angle, slow the motion, change the lighting word in setting — and regenerate.

This single habit saves more time than any other prompting tip.

You'll know it worked when you can point at the new clip and say which slot produced the difference you see.

Step 6 — Save the prompt that worked

Once you get a clip you like, copy the full prompt into a notes file or a spreadsheet. Name it after the scene ("coffee-barista-v3"). The next time you need a similar clip, you won't be starting from a blank text box again — you'll be tweaking a winner.

You'll know it worked when you have a small library of prompts you can adapt in under a minute.

Common mistakes

• Mistake: Describing the whole scene in one long run-on sentence. Fix: Break it into the six slots. Models respond better to comma-separated chunks than to prose paragraphs.
• Mistake: Asking for two different camera styles in one prompt. Fix: Pick one. "Tracking shot from the side AND overhead close-up" usually gives you neither.
• Mistake: Skipping motion entirely. Fix: Even "slow, steady motion" or "static frame" is better than nothing — without it, the model guesses and often guesses wrong.
• Mistake: Rewriting the whole prompt after a bad result. Fix: Change one slot, regenerate, then judge. You're trying to isolate which variable mattered.

Wrap-up

A six-slot prompt won't read your mind, but it comes close. Write the subject, action, setting, camera, motion, and audio (when supported), generate, watch, then change one slot at a time. Try one of the filled-in examples above today — pick the coffee barista, the child with dandelion seeds, or whichever scene caught your eye — and swap just one slot on your second try. You'll see the difference on the second clip, not the tenth.