AI Video Prompts Guide: Formula, Tips and 15 Examples

Good AI video prompts read like a director’s note for one shot: who is in frame, what they do, where it happens, how the camera moves, and what the light and sound feel like. Write that in plain sentences, keep it to one main action per clip, and most current video models will follow it far more closely than a pile of keywords.

Quick answer: the AI video prompt formula

• Shot + subject + action + setting + camera + light/mood, then optional sound and style.
• One clear action and one camera move per clip.
• Describe what you want to see, not what you want to avoid.
• For image-to-video, describe only the motion. The image already shows what things look like.
• Change one thing at a time when a result misses.

Below you’ll find the formula broken down, 15 copy-ready prompt examples across common genres, how prompting differs between text-to-video and image-to-video, model-specific tips for Seedance, Kling and Veo, and a fix-it table for the problems everyone runs into. Every tip is checked against the model makers’ own guides as of October 17, 2026.

The AI video prompt formula that works across models

Every major model maker now publishes a prompt guide, and they agree more than they differ. Kling’s prompt guide (August 2026) lists subject, action, scene, camera, and lighting and mood. Runway’s prompting guide boils it down to camera movement, scene, action and details. Google’s Veo prompt guide (updated October 15, 2026) breaks prompts into subject, action, scene or context, camera angle and motion, composition, ambiance, and audio.

Put together, a reliable order looks like this:

[Shot size and camera move]. [Subject with 2–3 visible details] [does one clear action] in [setting, time of day]. [Lighting and mood]. [Optional: sound]. [Optional: style or film look].

Lead with the shot. Runway’s guide notes that putting camera information first helps the model frame everything else. If you’re fuzzy on shot names, our guide to the types of camera shots covers every framing term these models understand.

15 AI video prompt examples you can copy

These are written to be pasted as-is, then adjusted. Each one keeps to a single action and a single camera move, which is the most common difference between prompts that work and prompts that fall apart. None of these were generated for this article, so treat them as starting points and expect to iterate.

Cinematic and narrative shots

Slow push-in, medium close-up. A woman in a yellow raincoat stands under a flickering neon sign on a narrow street at night, rain dripping from her hood. She lowers her phone and looks up, breath visible in the cold air. Pink and teal neon reflects on the wet pavement. Quiet, tense mood. Sound: steady rain, a distant siren, no music.

Aerial drone shot slowly gliding forward over a coastal cliff road at golden hour. A single red car winds along the road far below. Long shadows, warm orange light, sea mist rolling in from the right. Calm, wide, hopeful mood.

Slow dolly-in through a dark doorway into a candlelit study. An old man sits at a desk covered in maps, turning a brass compass in his fingers. Warm flickering candlelight, deep shadows, dust floating in the air. Soft crackle of candles, ticking clock.

Wide tracking shot running alongside a motorbike as it speeds across a cracked desert flat at noon. Dust plumes billow behind the rear wheel. Harsh overhead sun, heat shimmer on the horizon. Fast, gritty, high-energy mood.

Static close-up, shallow depth of field. A young man sits on a train, his face lit by passing station lights. A tear runs down his cheek and he wipes it away quickly. Cool blue night tones with warm flashes of light. Muffled rhythm of the train on the tracks.

Product and ad shots

Slow orbit around a matte black wireless headphone set resting on a wet slate surface. Water droplets bead on the ear cups. Soft studio key light from the left, crisp rim light outlining the shape, dark background. Premium, minimal mood.

Macro close-up, locked-off camera. Golden honey pours slowly onto a stack of pancakes, pooling and dripping over the edges. Morning window light from behind, steam rising. Warm, cozy, appetizing mood.

High-speed slow-motion shot. A glass perfume bottle drops into still turquoise water, sending up a crown-shaped splash. Bright clean studio lighting, tiny bubbles catching the light. Elegant, fresh mood.

Social and short-form shots

Vertical 9:16, first-person point of view. Hands pour coffee into a white mug on a sunlit kitchen counter, then lift it toward the camera. Soft morning light through sheer curtains, plants on the windowsill. Relaxed, cozy mood. Sound: coffee pouring, birds outside.

Medium shot, eye level, locked-off camera. A woman in her thirties sits at a tidy desk facing the camera, warm practical lamp behind her. She leans in slightly and says: "Here’s the one mistake everyone makes with AI video." Natural room tone, no music, no subtitles.

Top-down close-up. A palette knife slowly spreads thick teal paint across a white canvas in one smooth stroke. Even soft lighting, rich texture visible. Calm, satisfying mood. Sound: the soft scrape of the knife.

Stylized and camera-move shots

Stop-motion claymation style. A small clay fox in a knitted scarf walks across a felt meadow toward a tiny wooden house, slightly jerky frame-by-frame movement. Warm practical light, handmade textures visible. Whimsical mood.

90s cel-shaded anime style. Slow pan across a rooftop at dusk as a girl with a school bag watches trains pass below. Orange and violet sky, power lines silhouetted, city lights flickering on. Nostalgic mood.

Dolly zoom on a man standing in a long hospital corridor: the camera moves back while zooming in, so the corridor stretches behind him. He stays perfectly still, eyes wide. Cold fluorescent light. Unsettling, dizzy mood.

FPV drone shot flying fast and low through an abandoned warehouse, weaving between steel columns toward a bright open doorway. Pigeons scatter, dust hangs in shafts of sunlight. Exhilarating mood.

Want to use these in a sequence instead of as one-offs? Plan the shots first. An animatic or simple shot list keeps your prompts consistent from clip to clip.

Image-to-video prompts: describe the motion, not the picture

The biggest mistake with image-to-video is re-describing the image. The model can already see the woman in the red coat. Repeating it wastes the prompt and can make the model fight the image. Use your words for what moves and how the camera moves.

The camera slowly pushes in. She turns her head toward the window and smiles slightly. Her hair moves gently in a breeze. Curtains sway in the background. Light shifts subtly as a cloud passes.

• Name motion for the subject, the background and the camera separately.
• Keep it physically plausible for the frame you uploaded. A seated character can’t sprint across a room in five seconds without warping.
• Match the style. A painted still animated with “photorealistic” language tends to drift.

Some tools also take an end frame, but workflows differ. If you chain clips, a simple approach is to export the last frame of one clip and use it as the start image for the next.

Camera movement and lighting words AI video models understand

Models respond best to standard film language. Runway’s guide recommends one primary camera movement per prompt, because several moves in one clip tend to produce chaotic results.

For lighting, name visible sources rather than adjectives: “golden hour backlight,” “single overhead fluorescent,” “neon reflecting on wet pavement,” “soft window light from the left.” Combine them with lens language like “shallow depth of field” or “wide-angle” when the look matters. Our roundup of cinematography techniques for beginners is a good source of vocabulary.

How to prompt sound and dialogue

Several current models generate audio along with the picture. Google’s Veo guide recommends describing audio in separate sentences, and splits it into sound effects, ambient noise and dialogue. Kling says its 3.0 series handles dialogue, sound effects and ambience in one generation and recommends clear speaker labels when more than one person talks.

Medium two-shot in a small diner booth at night. The older man slides a coffee cup across the table and says: "You were never supposed to find this." The younger woman stares at the cup. Sound: low hum of a refrigerator, rain on the window, no music, no subtitles.

• Put spoken lines in quotes after “says:” so the model knows exactly what to say.
• Keep lines short. A 5–10 second clip fits one or two sentences.
• Add “no subtitles” if you don’t want burned-in text. ByteDance’s Seedance 2.5 guide lists “no subtitles” and “no BGM” as supported negative constraints.

Audio support varies by model and by tool, so check whether the version you use actually outputs sound before you write a soundtrack into the prompt.

Model-specific prompt tips: Seedance, Kling and Veo

The formula transfers, but each model has quirks worth knowing.

Seedance 2.5

ByteDance’s prompt guide for Seedance 2.5 (which can make clips up to 30 seconds) suggests a base order of subject, location, event, style and camera movement. For longer clips, split the prompt into timestamps or “Shot 1, Shot 2” segments. When you use reference images or clips, name each one by upload order (“Image 1 is the main character”) instead of relying on text inside the image. More in our Seedance 2.5 guide.

0–4s: Wide shot. A lighthouse keeper climbs a spiral staircase, lantern in hand, wind howling outside.
4–8s: Medium shot. He reaches the top and opens the hatch; rain blasts in and his coat whips around him.
8–12s: Close-up. He squints into the storm and spots a small boat’s light far out at sea.
Consistent details: cold blue storm light, warm lantern glow, handheld feel. Sound: wind, rain, creaking metal, no music.

Kling

Kling’s guide stresses plain, readable scene direction over “secret formulas” and supports multi-shot prompts written as labeled shots. For the Kling 3.0 series, the same guide lists native audio in five languages (Chinese, English, Japanese, Korean and Spanish). Kling 4.0, which launched in October, names more; see our Kling 4.0 guide for what changed.

Veo

Google’s guide is the most detailed on audio, and it documents negative prompts as a separate tool: describe what you don’t want plainly (“urban background, man-made structures”) instead of writing “no” or “don’t” in the main prompt.

Common AI video prompt problems and how to fix them

When a clip misses, change one variable at a time (the camera move, then the light, then the action) so you learn what actually made the difference. Character consistency across many shots is its own skill; our guide to consistent AI characters covers it in depth.

Using these prompts in CinemaDrop

If you’d rather plan a whole story than prompt clip by clip, CinemaDrop lets you write a script, break it into a storyboard of scenes and shots, and generate an image or video for each shot. Characters, locations and props can be saved as Elements with up to four reference images each, and you tag them into storyboard shot prompts with @ (for example, @Maya), so you don’t have to retype the same description every time.

• Video models include Seedance 2.5, Kling 3.0, Veo 3.1 and Wan 2.7, among others.
• Image-to-video in CinemaDrop uses a start frame.
• The free Lite plan includes 50 one-time credits for images, scripts and storyboards but no video. Video starts on the Basic plan ($8 a month), and Seedance 2.5 needs Plus.
• Export is a ZIP of your storyboard assets.

New to all of this? Start with our beginner’s guide to AI video.

AI video prompts FAQ

How long should an AI video prompt be?

Usually 40 to 100 words for a 5–10 second clip: enough to cover shot, subject, action, setting, light and sound. Longer clips with several beats can use timestamped segments.

How do I prompt AI to make a video?

Describe one shot in plain sentences: the camera framing and move, the subject and one visible action, the setting, the lighting and mood, and any sound. Then generate, review and change one thing at a time.

Which AI is best for video prompts?

There’s no single winner. Most current models (Seedance, Kling, Veo, Runway and others) follow the same plain-language structure. Pick the model for the job, such as clip length, audio or style, and use the formula above.

Do the same prompts work across different video models?

Mostly yes. The core structure transfers. What changes is support for audio, dialogue languages, clip length, multi-shot prompts and negative prompts.

Should I use negative prompts?

Prefer positive descriptions in the main prompt. Use a separate negative-prompt field only where the tool offers one, and keep simple constraints like “no subtitles” or “no music.”

Can I use an AI video prompt generator?

Yes, as a starting point. Generators help with vocabulary, but you’ll still get better results by editing the output down to one clear action and one camera move.

The bottom line

The best AI video prompts aren’t clever. They’re specific. Lead with the shot, give the subject one visible action, name real light sources, put sound in its own sentence, and change one thing at a time. Save the prompts that work, and you’ll build a personal library that beats any list of “viral prompts.”