Ijharul Islam
To write prompts for AI video generators, describe one shot in this order: subject, action, setting, camera, lighting and style, then sound. Be specific about what moves and how, keep it to one scene per prompt, and describe what you want rather than what you don't. A clear 40 to 80 word prompt usually beats a long list of keywords.
This guide gives you that formula, the words models respond to, twelve example prompts you can adapt, and fixes for the problems you'll see most. The advice works in Veo, Kling, Runway, Luma and most other models.
The short answer
Write one shot per prompt: who or what is in it, what they do, where, how the camera frames and moves, the lighting and style, and the sound or dialogue. Use concrete words ("slow dolly in, golden hour, shallow depth of field") instead of vague ones ("cinematic, amazing"). Generate short drafts on a cheap model, then refine.
Key takeaways
- Formula: subject + action + setting + camera + lighting/style + sound.
- One shot per prompt. Several actions in a few seconds usually come out muddled.
- Name the camera move and lens: "slow push in", "handheld", "close-up", "wide shot".
- For audio models, write dialogue as Speaker says: "line", and list sound effects in their own sentence.
- Describe what you want instead of writing "no"; Google's Veo guide recommends listing unwanted elements separately.
Disclosure: we make QuickVeed and show how to do this in it, along with other ways. Details about other products come from their public pages, checked in October 2026 and linked at the end.
The AI video prompt formula
Google's Veo prompt guide breaks a good prompt into subject, action, scene, camera angle and movement, lens effects, visual style, pacing and audio. That maps to a simple order you can use with any model:
| Part | Question it answers | Example |
|---|---|---|
| Subject | Who or what? | A woman in her 30s in a green apron |
| Action | What happens, and how? | slowly pours latte art into a white cup |
| Setting | Where and when? | in a small sunlit cafe in the morning |
| Camera | Framing and movement? | close-up, slow push in, shallow depth of field |
| Lighting and style | What does it look like? | warm natural light, soft shadows, realistic |
| Sound | What do we hear? | quiet cafe chatter, milk steaming |
Put together: "Close-up of a woman in her 30s in a green apron slowly pouring latte art into a white cup in a small sunlit cafe in the morning. Slow push in, shallow depth of field, warm natural light, soft shadows, realistic. Quiet cafe chatter and the hiss of a milk steamer."
Words that give you control
Camera
- Framing: extreme close-up, close-up, medium shot, wide shot, aerial shot, over-the-shoulder.
- Angle: eye level, low angle, high angle, top-down.
- Movement: static, slow push in, pull back, pan left, tilt up, tracking shot, orbit, handheld.
- Lens: shallow depth of field, wide-angle, macro, rack focus.
Light and style
- Time and light: golden hour, overcast, neon at night, soft studio light, backlit.
- Look: photorealistic, documentary, 1990s camcorder, clay animation, flat 2D animation.
- Pace: slow motion, time-lapse, real time.
Sound (models with native audio)
Describe sound in separate sentences. For dialogue, name the speaker and use "says:", for example: The barista smiles and says: "Your usual?" Keep spoken lines short; a few seconds of clip only holds about one sentence.
12 example prompts
| Use | Prompt |
|---|---|
| Product hero | A matte black water bottle rotating slowly on a wet stone surface, water droplets on its sides. Macro close-up, slow orbit, dark background, single soft key light, photorealistic. |
| Food | Overhead shot of hands sprinkling sea salt on a sizzling steak in a cast-iron pan. Slow motion, warm kitchen light. Sizzling sound. |
| Real estate | Slow drone shot rising over a white two-storey house with a pool at golden hour, revealing a quiet suburban street. Realistic, soft warm light. |
| App promo | A smartphone on a desk lights up with a notification, a hand picks it up. Close-up, shallow depth of field, clean modern office, cool daylight. |
| Talking character | Medium shot of a friendly older man in a hardware store holding a hammer. He looks at the camera and says: "Pick the right tool, and the job takes half the time." Store ambience. |
| Travel | Handheld tracking shot following a cyclist through a narrow cobbled street in the morning, flower boxes on windows. Natural light, documentary style. Bicycle bell and birdsong. |
| Fitness | Low-angle shot of a woman finishing a box jump in a bright gym, chalk dust in the air. Slow motion, high contrast light. |
| Abstract background | Slow flowing ribbons of violet and teal light on a black background, smooth looping motion, soft glow. |
| Explainer B-roll | Top-down shot of a person writing a to-do list in a notebook next to a coffee cup, then ticking the first item. Soft daylight, calm. |
| Nature | Wide shot of fog rolling over pine-covered hills at sunrise, slow time-lapse, cool blue tones turning warm. |
| Animation | Flat 2D animation of a small paper boat sailing across a blue page, waves drawn in simple lines, pastel colours, gentle motion. |
| Image to video | (With a product photo uploaded) The camera slowly pushes in as soft light moves across the product. Keep the product unchanged. |
Fixing common problems
| Problem | Likely cause | Fix |
|---|---|---|
| Nothing much happens | No clear action | Add a verb and a speed: "slowly pours", "turns to camera" |
| Chaotic or warped motion | Too many actions for the length | One action per clip; split the rest into new shots |
| Wrong look | Vague style words | Replace "cinematic" with lighting, lens and colour words |
| Unwanted objects or text | Model filling space | Describe the background exactly; list unwanted items separately if the model supports a negative prompt |
| Hands or faces look wrong | Complex close-ups | Use a wider shot, or start from a good image |
| Product changes shape | Text-only prompt | Upload the product photo as the first frame |
| Dialogue is garbled | Line too long for the clip | Shorten to one sentence, or pick a longer clip |
Google's guide advises against writing "no" or "don't" in the prompt itself, because models can latch onto the word that follows. Describe the scene you want instead.
Prompting for image-to-video and for dialogue
When you start from an image
If you upload an image, the model already knows what the scene looks like. Describe only what should move and how the camera behaves. Repeating the contents of the picture can make the model change them. Good image prompts are short: "Slow push in, steam rises from the cup, light flickers through the window." If you want the product or face to stay exactly the same, say so: "Keep the bottle unchanged."
When the model speaks
Models with native audio can generate a character saying a line. A few rules make it work:
- One short sentence per clip. A 6 to 8 second clip holds about 12 to 18 spoken words.
- Name who speaks, so the model knows which mouth to move: The woman at the counter says: "..."
- Describe the voice if it matters: calm, cheerful, a whisper.
- Put sound effects and background in their own sentence, after the dialogue.
- Avoid quotation marks inside the line, and spell out numbers.
Keeping a series consistent
For several clips with the same character or product, write a fixed description once ("a woman in her 30s with short black hair, a green apron and round glasses") and paste it unchanged into every prompt. Better still, start each clip from the same reference image. Change only the action and the camera between clips.
Testing prompts without wasting credits
- Draft on a cheap, short model to check composition and motion.
- Change one thing at a time: the camera, then the light, then the action.
- Once the shot works, render it on the model you want for the final, at the length and resolution you need.
- Save prompts that work in a document, organised by shot type.
In QuickVeed's AI video generator you choose from 15+ models and see the price before generating, so drafting is cheap: a 2-second Vidu Q3 Turbo clip at 540p is 20 credits and a 5-second Luma Ray 3.2 clip is 38, against 300 for an 8-second Veo 3.1 Fast clip with audio. Failed generations are refunded. To choose between the top models, see Veo 3.1 vs Kling v3.
Frequently asked questions
How long should an AI video prompt be?
Usually 40 to 80 words: enough to cover subject, action, setting, camera, style and sound. Very long prompts often have details the model ignores.
What makes a good AI video prompt?
One clear shot, a specific action with a speed, concrete camera and lighting words, and a short description of the sound if the model generates audio.
Do the same prompts work in Veo, Kling and Runway?
Mostly, yes. The formula works across models, but each responds a little differently, so test the same prompt on two models before a big batch.
How do I make AI video characters talk?
Use a model with native audio, such as Veo 3.1 or Kling v3, and write the line as: The man says: "...". Keep it to one short sentence per clip.
Should I use negative prompts?
Describe what you want first. If a model has a separate negative prompt field, list the unwanted things there as plain nouns, not as "no" or "don't".
Try it with your own video
Paste one of the example prompts and try it on a cheap model first. Describe the video or paste your script, pick a voice, and judge the first draft for yourself. QuickVeed plans start at $18/month, and new accounts get a 14-day trial with 75 credits, no card needed. Sign up or see pricing.
Sources
Competitor details in this post were checked against these public pages in October 2026. Products and prices change, so confirm with each vendor before you decide.
