The AI video prompt examples that work describe one shot the way you would brief a camera operator: who or what is on screen, what it does, where, how the camera moves, the light, the look and the sound. Copy one that matches your job, swap in your subject, and change one part at a time.
AI video prompt examples you can copy
Each prompt describes a single shot with one action and one camera move. Swap in your subject and setting and keep the order. Length, aspect ratio and resolution are usually generator settings, so set them there, not in the prompt.
Product shot on a clean surface
A matte white skincare bottle with a silver cap stands on a pale stone slab. The camera moves in a slow half orbit from left to right at label height. Soft window light from the left, a gentle shadow to the right, plain warm gray background. The bottle and its label stay perfectly still and sharp. Quiet room tone.
The half orbit adds motion without asking the product to move. The last visual line names what must not change: the bottle and its label.
Product in use
Medium shot of a woman in gym clothes standing at the open trunk of a car in a gym parking lot, early morning. She twists the lid onto a portable blender and gives it one shake. Handheld phone footage, natural daylight, slight camera wobble. Sounds: the lid clicking shut, liquid sloshing, distant traffic.
The setting carries the idea, a blender used outside a kitchen, so the shot is wide enough to show the car. The action is two beats, a twist and a shake, which suits a short clip.
Vertical social clip with a hook in the first second
A barista slides a tall iced coffee across a wooden counter straight toward the camera and it stops just in front of the lens. Low angle, static camera, shallow depth of field, bright cafe behind in soft focus. Vertical framing with empty space in the top third for a caption. Sound: glass sliding on wood, ice clinking.
Movement toward the lens puts the hook at the start. Set the 9:16 aspect ratio in the generator. The empty top third is for a caption you add in the edit, since generated text is often garbled.
Cinematic scene
Wide shot of a lone hiker in a red jacket walking along a ridge at dawn, mist filling the valley on both sides. The camera tracks slowly alongside at a distance. Cold blue light with a warm rim of sun behind the hiker, anamorphic look, light film grain. Sound: wind, boots on gravel.
Shot size, one camera move and one lighting recipe do the cinematic work. Repeat the red jacket in later shots to keep the character recognizable.
B-roll for a voiceover
Overhead shot of a tidy desk: an open laptop, a notebook, a cup of coffee with steam rising. A hand enters from the right and writes one line in the notebook. Static camera, soft morning light from a window at the top of frame, muted natural colors. Ambient sound only: pen on paper, a quiet room.
B-roll sits under narration, so the camera is locked, the audio is ambient and one small action keeps the shot alive.
What are some good AI video prompts?
Good AI video prompts name a specific subject, give it one clear action, and state the camera, light and sound instead of leaving them to chance. They read like a shot description, not a list of adjectives. Google's video generation prompt guide for Veo says the same of the subject: specificity helps avoid generic outputs.
Explainer scene
Flat 2D animation in a limited palette of navy, coral and cream. A small parcel travels along a dotted line from a warehouse icon on the left to a house icon on the right, pausing once at a checkpoint in the middle. Static camera, smooth even motion, plain cream background, clean shapes.
An explainer needs one idea per scene, so the prompt describes a single journey. The palette is named so every scene can share it.
Ad opening
A man in a rain-soaked coat stands under a bus shelter at night and unscrews a small silver thermos. Steam rises into the cold air and he smiles. Medium shot pushing in slowly to a close-up of the thermos. Wet street reflections, neon signs out of focus behind. Sound: rain on the shelter roof, a soft exhale.
The problem (cold, wet) and the payoff (warmth) are both on screen. The push-in ends on the product, so the last frame can serve as a pack shot.
How to write a good prompt for an AI video?
Write the elements in this order: subject, action, setting, camera, lighting, style, audio. Put the most important thing first and describe one shot, with one camera move and one main action. Each element you leave out is a decision the model makes for you.
Element | What it decides | Sample phrase |
|---|---|---|
Subject | Who or what the shot is about | a ceramic mug with a chipped blue glaze |
Action | What happens, in one or two beats | is filled with coffee, steam rising |
Setting | Place and time of day | on a kitchen windowsill at sunrise |
Camera | Shot size, angle and movement | close-up, eye level, slow push-in |
Lighting | Source, direction and mood | warm backlight, long soft shadows |
Style | The look of the footage | 35mm film, shallow depth of field |
Audio | Speech, effects, ambience | pouring liquid, birdsong outside |
How long should an AI video prompt be?
Three to six sentences covers most single shots. OpenAI's Sora 2 prompting guide describes the trade-off: shorter prompts give the model more creative freedom, and longer, more detailed prompts restrict it. It also notes that models follow instructions more reliably in shorter clips, so consider two short shots instead of one long one.
Can a prompt include dialogue and sound?
Yes, on models that generate audio. Google's guide recommends describing audio in separate sentences and splits it into sound effects, ambient noise and dialogue, with speech written as "the man in the red hat says: Where is the rabbit?". Keep spoken lines short.
What to leave out
If your generator has a negative prompt field, list the unwanted things as plain nouns, such as "text, watermark, extra hands". Google's guide advises against instructive wording like "no" or "don't" in that field. Luma's guide to prompts for video ads recommends exclusions to avoid distorted labels, fake logos and unreadable text.
AI video prompt template
Fill in each bracket with one concrete phrase and delete any line you do not need. The last line, borrowed from the structure in Kling's prompt examples, says what must stay still.
Worked example built from the template
Medium close-up at eye level of a baker with flour on her forearms and a green apron, scoring a round loaf with a blade in a small bakery before sunrise. The camera stays static. Warm light from the oven on the right, dark blue window behind, documentary style. Sound: the blade on the crust, a low oven hum. Her face and apron stay the same throughout.
Each sentence maps to one template line, so you can change the camera or the light alone.
Text to video and image to video prompt examples
A text-to-video prompt has to describe everything. An image-to-video prompt describes only what changes, because the picture already fixes the subject, setting and style. OpenAI's guide puts it this way: the image anchors the first frame, and the text prompt defines what happens next.
Text to video: the whole scene in words
A wide shot of a white sailboat with a single mast crossing a quiet bay at sunrise. Calm water, low orange sun behind the boat, thin mist near the shore. The camera drifts slowly to the right, keeping the boat in the center. Sound: water against the hull, a distant gull.
Image to video: the same idea from a still
The sailboat holds its course and glides forward as morning light spreads across the bay. Gentle ripples, the sail shifts slightly in the wind. The camera drifts slowly to the right.
The still supplies the boat and the bay, so the prompt is only motion: what moves, how much, and what the camera does.
In the Protoface docs, the still is supplied as the opening frame of the clip, and the field that carries it differs from model to model. Start from a sharp image at the aspect ratio you will publish. A real product photo keeps the product accurate, and a frame from your footage helps generated shots match filmed ones when you plan a hybrid AI and live-action shoot.
Prompt examples for talking presenters and avatars
A presenter prompt needs four things: who is speaking, the framing, the delivery, and the exact line. Keep the camera still and the line short. OpenAI's guide suggests one or two short exchanges for a four-second shot.
Scripted presenter to camera
Medium shot of a woman in her thirties with short dark hair and a navy blazer, standing in a bright modern office, looking directly into the camera. Static camera at eye level, soft key light from the front left. She speaks in a friendly, steady tone and says: Your first report is ready in two minutes. Quiet office ambience.
Eye line, framing and tone make the clip read as a presenter. The line is eight words, which leaves room for a natural pause.
Selfie-style testimonial
A man in his twenties films himself with a phone held at arm's length in a small kitchen, daylight from a window to his left. Slight handheld movement, casual clothes. He smiles and says: I stopped skipping breakfast the week I got this. Natural room sound.
The handheld selfie framing sets the user-generated look. Do not present a generated person as a real customer: use the clip as a storyboard for a real testimonial.
A voiceover works the same way: write "a calm female voiceover says:" and the line, so the model knows nobody on screen is speaking. An avatar that answers people live is a different product: Protoface Realtime gives a voice agent a face driven by the audio the agent already produces.
What are some trending AI video prompts?
Trending prompts are formats more than subjects, and they turn over quickly, so treat any list as dated. Two formats listed in YouMind's public video prompt library in October 2026 were the camcorder-style vlog and the point-of-view shot.
Camcorder-style vlog
Home video from an old handheld camcorder: a teenager shows the camera around a garage band rehearsal, panning quickly between a drummer and a guitarist. Soft washed-out color, visible tape noise, slight zoom hunting. Sound: muffled drums, laughter, a guitar being tuned.
The look comes from naming the recording device and its flaws: tape noise, washed color and an unsteady zoom.
Point-of-view shot
First-person point of view of two hands opening a cardboard box on a doorstep, lifting out a pair of white sneakers wrapped in tissue paper. The camera looks down as the viewer would, with small natural head movement. Overcast daylight. Sound: cardboard tearing, paper rustling.
Point of view puts the viewer in the scene, so the prompt describes hands and a downward look, not a person seen from outside.
Start with what the video has to prove
Before you write a prompt, write one line that says what the clip should make a viewer do or believe. That line is your creative hypothesis, and the prompt is one way of filming it.
Use this form: "Showing [creative choice] will [viewer response] among [audience]." For example: "Showing the blender in a car trunk after a workout will hold first-time viewers past the opening seconds." The product-in-use prompt was written from that sentence.
Change one variable per version. Keep the product, length, framing and call to action fixed, and change only the opening, the setting or the emotion.
Vary the idea, not the wallpaper. A new actor or color grade is the same concept. A new concept changes the viewer's reason to care: saving time, or a smaller grocery bill.
Label every clip with its hypothesis and the variable it changes.
The same applies to a pitch for something that does not exist yet: decide what the audience must believe first, as covered in when to make a concept trailer.
Why a prompt fails and how to fix it
Most failed clips trace back to one of four causes: a vague subject, too many actions, nothing said about what must stay the same, or text the model was asked to draw.
Problem | Before | After |
|---|---|---|
Generic output | A person drinking coffee in a cafe | An elderly man in a tweed cap stirs an espresso at a marble counter, close-up, morning light from the left |
Chaotic motion | She runs in, grabs the bag, jumps the fence and waves | She lifts the bag from the bench and turns to the camera. Static shot |
Product changes shape | A handbag carried down a street | Start from a product photo. The bag swings gently at her side and its shape, color and handle stay unchanged |
Garbled text | A sign that says Grand Opening Sale | A blank white sign above the door. Add the words in your editor |
Why does the same prompt give different results each time?
Generation is not deterministic. OpenAI's guide states that using the same prompt multiple times leads to different results and advises treating the prompt as a wish list, not a contract.
Drifting characters and objects
Temporal consistency is how well a face, an object and the lighting stay the same from the first frame to the last. To protect it, shorten the clip, simplify the action, repeat the same identifying details in every prompt, and start from a reference image when identity matters.
Review the clip before you publish it
Score each clip against the job it has to do, not against "looks good".
Prompt adherence. List the instructions that must appear on screen and tick each one.
Subject fidelity. Compare the product or person with your reference image.
Motion. Watch at normal speed, then step through the fastest moment looking at hands, edges and shadows.
Usability. Check room for captions, a clean start and end, and safe framing for your crop.
Record the reason a clip was rejected ("wrong product", "motion breaks at the turn"). Those notes tell you what to change in the next prompt.
Delivery checks
Frame rate. It matters most in the edit. A clip that does not match the timeline's frame rate gets repeated or blended frames, which can show as stutter in fast motion. Check the rate your generator outputs and set the timeline to it.
Platform compression. Social platforms re-encode uploads. Fine texture, small text and soft gradients tend to suffer most, so check the uploaded version on a phone.
App Store previews. Apple's app preview rules require footage captured from the app itself. Keep generated lifestyle scenes for the ads that promote the app.
Rights and handoff. Get approval for any real likeness, brand mark or source image before paid use, and send the editor the prompt, model, settings and known flaws with the file.
Every rejected take uses generation allowance, so set a number of takes per shot before you start. See AI video generator quotas, credits and budgets. To try any of these prompts, open Protoface Studio, paste one in and change a single element between takes.
Common questions
Do image-to-video prompts need different details from text-to-video prompts?
Yes. The image already fixes the subject, setting and style, so the prompt should describe only the motion: what moves, how much, and what the camera does. A text-to-video prompt has to describe the whole scene.
Does one prompt work across different video models?
The description of the shot carries over. The settings do not: models differ in clip length, aspect ratios, whether they generate sound and how they take a starting image. Keep the prompt and adjust the settings for each model.
Is a free AI video prompt generator worth using?
It is useful for a first draft, because it reminds you to state the camera, light and sound. It cannot know what your video has to prove or what your product looks like, so edit its output: replace generic subjects with your own details and cut every adjective that does not change the picture.
What should you do when a generation fails?
Work out which kind of failure it is. A technical error can simply be run again. A clip with broken motion needs a shorter, simpler action. A polished clip that shows the wrong thing needs a more specific subject or a reference image, not another identical attempt.
Should you score AI video by shot or by finished video?
Both. Review each shot before editing so weak footage is replaced while that is still quick, then review the finished video for message, pacing and fit with the placement, which no single shot can show.
Try these prompts on your own product
Open Protoface Studio, paste a prompt, and change one element between takes to see what each part of the prompt controls.
