AI WORLD HQ
You've opened an AI video tool, typed "a cat playing piano," and watched it make you a 10-second clip with random music you didn't ask for and a sudden camera zoom you didn't want. The result isn't bad — but it isn't what you pictured either. The fix usually isn't the tool. It's the prompt. (A prompt is just the instruction you type to an AI — think of it as your creative brief.)
Newer video models from OpenAI, Google, Runway, and Kuaishou can do something older tools couldn't: make the picture and the sound in the same pass. That means silence is something you have to ask for, dialogue is something you quote, and on-screen text is something you write exactly the way you want it to appear. The six prompt templates below work across most of these tools.
Before you start
You don't need anything beyond what you already use to type into an AI video tool. The patterns below assume your tool accepts text prompts (most do). If your tool has a specific "include sound" or "audio generation" toggle in its settings, turn it on — but the prompt patterns themselves are the same whether sound is on or off. For exact menu names and pricing, check your tool's official help page, because each tool labels these options differently.
Step 1 — Set the scene first
Start every prompt with where the scene happens, when (time of day or mood), and the overall feeling you want. Without this, the AI picks for you — and you usually get something generic.
A scene line should be one short sentence at the very top of your prompt, before any action.
💬 A quiet beach at sunset, warm orange light, gentle waves.
You'll know it worked when the AI's first shot clearly shows that beach at sunset, not a different beach at noon.
Step 2 — Describe the action and the camera
After the scene, add what happens — and what the camera does. Action plus camera movement is the part most beginners skip, which is why their clips feel static.
Try one sentence for action and one for the camera:
💬 A woman walks slowly along the shoreline, pausing to look at the horizon. Slow dolly shot moving with her. ("Dolly" means the camera moves physically toward or away from the subject, like on a track.)
You'll know it worked when the clip actually moves — there should be at least one camera move, not just a still image.
Step 3 — Specify sound (or ask for silence)
This is the part that catches people off guard with newer tools. If you don't mention audio, the AI will guess — and add music, narration, or random sound effects you didn't want.
You have three options, and you should pick one every single time:
- Sound on, no dialogue: name the sounds you want.
- Dialogue: see Step 4.
- Silence: explicitly say so.
💬 Only the sound of waves and a single seagull in the distance. No music.
You'll know it worked when the clip has clear audio that matches what you described, or no audio at all if you asked for silence.
Step 4 — Quote dialogue instead of describing it
If you want a character to speak, write their exact line in quotes. AI video tools render speech much better when you give them the words than when you describe what someone should "say."
💬 The woman turns to camera and says, in a soft voice: "I've been waiting here for an hour."
Avoid:
- ❌ "She says something romantic about the sunset" — too vague.
- ✅ "She says: 'This is my favorite place.'" — exact words, in quotes.
You'll know it worked when the spoken line in the clip matches (or closely resembles) what you wrote. Lip-sync (matching mouth movement to the words) is still imperfect, so aim for short, slow lines.
Step 5 — Quote on-screen text exactly
If you want a title card, a label, or a caption, write the text in quotation marks with a short note about font style and timing. Modern tools can render legible text, but only if you tell them what the text is.
💬 The video opens on a black screen with the text "CHAPTER ONE" in white serif font, fading in slowly over two seconds. ("Serif" means letters with small decorative tails, like the fonts in most printed books.)
You'll know it worked when the words appear on screen in a readable form, not as garbled letters. Keep on-screen text short — a few words is far more reliable than a full sentence.
Step 6 — Change one thing at a time when iterating
Your first clip almost never matches your idea. That's normal. The trick is to edit one part of the prompt at a time, so you learn what each piece controls.
A useful workflow:
- Save each prompt and the resulting clip somewhere you can find later.
- Change one element (camera angle, time of day, sound, dialogue) and re-run.
- Compare results side by side.
You'll know it worked when you can point to a specific prompt change that produced a specific improvement — not when you've changed five things at once and don't know what helped.
Common mistakes to avoid
Mistake: A vague opening. "A cool video of a city." — The AI has nothing to anchor on. Fix: Add location, time, and mood in your first sentence.
Mistake: Letting the AI decide on sound. You'll get random music over everything. Fix: Always end with one of three options: describe sounds, quote dialogue, or say "silent."
Mistake: Forgetting on-screen text. If you want words on screen, the AI won't add them by itself. Fix: Quote the exact text you want, with a brief note on font and timing. Keep it short.
Mistake: Editing five things at once. Then you don't know what helped. Fix: Change one element per try and save the prompt each time.
Mistake: Long dialogue lines. The AI stumbles on full sentences spoken in 10 seconds. Fix: Keep spoken lines to a short phrase, around 8-12 words.
Wrap-up
The biggest shift in newer AI video tools is that you now write the sound as carefully as the picture. Six prompt patterns will cover most of what you want to make: scene-first setup, action with camera moves, deliberate sound, quoted dialogue, quoted on-screen text, and one-change-at-a-time iteration. Pick one prompt from this article, paste it into your tool, and change only the scene line first. That's the smallest possible experiment, and it's the one that teaches you the fastest.