50 AI Video Prompt Examples: Cinematic, Product, Social, Dialogue, and Reference

50 AI Video Prompt Examples for Cinematic, Product, Social, Dialogue, and Reference Videos

AI video prompt examples work best when they describe one clear shot instead of an entire film. Start with the subject, then add the action, setting, camera movement, lighting, visual style, timing, and anything that must remain stable. The 50 examples below are ready to copy, but they are more useful as adaptable templates: replace the subject, brand details, location, aspect ratio, or motion while keeping the shot direction specific.

For text-to-video, describe everything the model must invent. For image-to-video, focus on what should move, what should stay unchanged, and how the camera should behave. Generate a short draft first, review subject consistency and motion, then revise one variable at a time. This process works for cinematic scenes, product videos, travel clips, social ads, dialogue scenes, and reference-driven animation. Every prompt below has a "Try this prompt" link that opens it in the generator with the text already filled in. If you need a broader prompting foundation first, read the Text to Video Prompt Guide.

We also rendered 10 of the examples on Seedance 2.5 and put each 4-second render under its prompt, together with what the model did not follow. Start with how we tested 10 of these prompts if you want the patterns first.

Prompt examples help creators move from a blank page to usable video ideas.

TL;DR

• Write prompts as shot directions: subject + action + setting + camera + lighting + style + constraints.
• Keep each generation focused on one scene and one main action.
• For image-to-video, describe motion and preservation instead of redescribing the entire image.
• Draft with simple motion first; add camera moves and atmosphere only after the subject behaves correctly.
• Use the examples below as structures, not magic phrases.
• In our 10 test renders, single-action prompts landed best. Unnamed labels and signs came back as made-up lettering, and requested orbits and slow push-ins were often skipped.
• To test a prompt right away, the AI video prompt library and the Seedance 2.5 prompt library open each example in the generator with one click.

Quick Answer

A strong AI video prompt answers seven questions: Who or what is on screen? What happens? Where does it happen? How does the camera move? How is the scene lit? What should it feel like? What must not change? A practical formula is:

[Subject] + [action] + [environment] + [camera] + [lighting] + [style and pacing] + [constraints]

For example: “A ceramic perfume bottle rotates slowly on wet black stone, soft mist moving behind it, macro dolly-in, narrow rim light, premium editorial product film, controlled reflections, label remains sharp and unchanged.”

Quick Prompt Formula

A reusable formula for writing better AI video prompts.

Last verified: October 11, 2026, against the SeedVideo AI workspace panels for Seedance 2.5, Kling 3.0, Veo 3.1, and Wan 3.0. Updated October 14, 2026: 10 examples rendered on Seedance 2.5, with results under each tested prompt.

Before using a long prompt, decide which element matters most. If product accuracy is critical, prioritize constraints. If the goal is atmosphere, prioritize lighting, camera, and environmental motion. If the scene starts from a reference image, use the workflow in the Image to Video AI Guide.

How the same prompt changes by model

The prompt structure above works everywhere, but the panel you type it into does not. These are the differences you hit most often when you move one prompt between the four video models in the SeedVideo AI workspace, as checked on October 11, 2026.

Two habits transfer well: keep one lead subject and one main camera move per clip, and put the non-negotiable constraint on its own short line so it is not the first thing a model drops. Separate negative-prompt fields are not something to assume; if a panel does not show one, write the constraint in ordinary prompt language.

Prompt length by clip duration

Longer clips need more structure, not more adjectives. These are editorial recommendations from writing and revising the prompts on this page:

How we tested 10 of these prompts

On October 14, 2026, we rendered 10 of the 50 prompts once each on Seedance 2.5, copied word for word from this page: two from each of the first five groups (cinematic, product, travel, social, and image-to-video). We did not retry or edit any prompt, so each clip is a first attempt.

• Settings: Seedance 2.5, 480p, 4 seconds, audio off, one render per prompt. Text prompts used 16:9 unless they ask for vertical 9:16. The two image-to-video prompts used the source image as the first frame with an adaptive ratio.
• Inputs: the portrait and the serum bottle used for examples 25 and 26 are AI-generated stand-ins, not real people or products.
• Generation time: measured from submission to finished clip. Your wait will vary with the queue and settings.
• Length: 4 seconds is the shortest clip the model accepts. Most of these prompts describe one action, so 4 seconds shows whether that action lands; a longer clip can still behave differently.

What the 10 renders had in common:

• One main action lands; the last step of a chain does not. The shoe strike (8) and the room split (20) followed almost every word. Prompts that end with a second beat lost it: the detective (1) never looked down the street.
• Text you do not spell out gets invented. The skincare label (7) and the market signs (14) filled with lettering that reads as nothing. The serum bottle (26) kept its real "SERUM" label because it came from the source image. If text matters, give the exact words or ask for a blank label.
• Camera moves are the first thing dropped. The macro orbit in 7 never happened, and the push-ins in 25 and 1 were slight. Give the move its own short sentence, or test it on a locked subject first.
• Image-to-video kept identity but not numbers. The portrait (25) and the bottle (26) kept face, shirt, shape, and cap. The "20-degree turn" in 26 became a turn of well over 45 degrees, so treat angles as a direction rather than a measurement.

Each tested example below ends with a Test result note and the render itself as a 4-second clip, looped without sound.

Cinematic AI Video Prompts

When to use them: narrative openings, concept trailers, music videos, mood films, dramatic transitions, and visual pitch decks.

1. Rainy detective entrance

A tired private detective in a dark wool coat steps out of a narrow hotel doorway into heavy midnight rain, pauses under a flickering red sign, then looks down the empty street; slow low-angle dolly-in, wet pavement reflections, deep shadows, restrained neo-noir color, realistic rain and coat movement, no sudden camera shake.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 16:9, 480p, 4 seconds, audio off, generated in 2 min 24 s): The detective in a dark coat steps out of a doorway under a red neon sign into heavy rain, with wet pavement, deep shadows, a red-and-teal noir palette, a slow push-in, and no camera shake. Did not follow: he ends facing the camera instead of looking down the empty street, and the doorway reads as a back-alley door rather than a hotel entrance.

Example 1 test render, the full 4-second clip (looped).

2. Desert discovery

A lone archaeologist climbs the final ridge of a vast desert dune and discovers a partially buried stone observatory below; begin behind the character, then crane upward into a wide reveal, early sunrise haze, wind lifting sand from the ridge, epic scale with grounded documentary realism, slow deliberate pacing.

Try this prompt in SeedVideo AI

3. Quiet science-fiction corridor

An engineer walks through a silent orbital-station corridor while emergency lights activate one section at a time ahead of her; centered tracking shot, subtle handheld tension, cool metallic surfaces, soft red warning light, realistic walking pace, consistent costume and face, no monsters or extra characters.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 16:9, 480p, 4 seconds, audio off, generated in 2 min 10 s): The engineer walks toward the camera down a centered metal corridor, red warning light comes on, and her face and dark jumpsuit stay the same with no extra characters. Did not follow: the red light washes over the whole corridor at once and then pulses along the walls, instead of switching on one section at a time ahead of her.

Example 3 test render, the full 4-second clip (looped).

4. Forest memory scene

A child in a yellow raincoat runs through a misty pine forest and stops beside an old wooden swing moving by itself; medium follow shot transitioning to a slow circular move, diffuse morning light, muted green palette, dreamlike but photorealistic, gentle fog drift, emotionally quiet rather than frightening.

Try this prompt in SeedVideo AI

5. Rooftop decision

Two friends stand on a city rooftop just before sunrise, one hands the other a sealed letter, both remain silent as traffic glows far below; slow side-profile push-in, natural wind in hair and clothing, cool blue hour shifting toward warm horizon light, intimate cinematic drama, subtle expressions.

Try this prompt in SeedVideo AI

6. Historical market reveal

A merchant opens a carved wooden shop at dawn in a crowded ancient coastal market; start on a close-up of the key turning, pull back through hanging fabrics to reveal vendors, baskets, and ships beyond, warm directional sunlight, textured period realism, layered background motion without chaotic crowd behavior.

Try this prompt in SeedVideo AI

Product Video Prompts

When to use them: ecommerce hero clips, launch videos, paid ads, product-detail pages, pitch decks, and visual prototypes. Replace generic product descriptions with accurate materials, colors, and packaging details.

7. Premium skincare bottle

A frosted glass skincare bottle stands on pale travertine as a thin ribbon of water flows around its base; slow macro orbit, soft morning side light, clean cream background, premium beauty commercial, controlled highlights, bottle shape and label remain unchanged and readable, no hands.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 16:9, 480p, 4 seconds, audio off, generated in 2 min 29 s): A frosted glass bottle with a pale cap stands on a travertine block against a cream background in soft side light, with no hands. Did not follow: the water sits as a still puddle around the base instead of flowing, the camera stays locked instead of orbiting, and the label is filled with made-up lettering because the prompt asked for a readable label without saying what it reads.

Example 7 test render, the full 4-second clip (looped): the bottle holds still and so does the water.

8. Athletic shoe performance shot

A lightweight running shoe lands on a wet track in slow motion, compresses naturally, then pushes forward as droplets scatter; low side angle, short tracking move, crisp stadium rim light, energetic sports-ad pacing, accurate sole geometry, no floating parts or distorted laces.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 16:9, 480p, 4 seconds, audio off, generated in 2 min 17 s): One of the two closest matches in the set. The shoe lands on a wet track, flexes, and pushes off as droplets spray, seen from a low side angle with rim light, and the sole and laces stay intact. Did not follow: there is no stadium; the background is a plain gray haze.

Example 8 test render, the full 4-second clip (looped).

9. Coffee maker demonstration

A compact stainless-steel coffee maker brews one espresso in a modern kitchen, steam rises as the cup fills, then the camera moves toward the crema; three-quarter product view into a gentle push-in, warm window light, realistic appliance operation, clean lifestyle commercial, no extra buttons or changing logo.

Try this prompt in SeedVideo AI

10. Headphones color launch

Matte cobalt over-ear headphones rotate slowly above a matching geometric pedestal while narrow bands of light sweep across the ear cups; locked centered composition with subtle zoom, dark blue studio, precise premium technology aesthetic, smooth controlled rotation, product proportions stay consistent.

Try this prompt in SeedVideo AI

11. Watch detail sequence

A brushed-steel wristwatch rests on black fabric as the second hand moves and the fabric folds shift slightly; extreme macro close-up transitioning to a top-down hero frame, focused specular lighting, luxury editorial style, dial markings remain sharp, no invented complications or changing time display.

Try this prompt in SeedVideo AI

12. Packaged snack social ad

A colorful bag of sea-salt chips drops onto a bright kitchen counter, three chips bounce beside it, then a lemon and herbs slide into frame; fast overhead-to-front transition, sunny hard light, playful commercial energy, packaging design remains stable and legible, no additional products.

Try this prompt in SeedVideo AI

Different video goals need different prompt structures.

Travel Video Prompts

When to use them: destination teasers, tourism campaigns, itinerary videos, hotel content, travel storytelling, and location mood boards. Avoid asking one clip to cover an entire trip; create one shot per moment.

13. Alpine lake arrival

A hiker reaches a clear alpine lake at sunrise, removes a backpack, and watches the first light touch the mountain peaks; wide rear three-quarter shot, slow forward glide, light fog above the water, natural documentary color, gentle clothing movement, realistic scale and terrain.

Try this prompt in SeedVideo AI

14. Night market walk-through

The camera moves slowly through a lively night market with steaming food stalls, handwritten signs, and small groups sharing dishes; eye-level stabilized walk-through, warm practical lights mixed with cool evening ambience, observational travel-film style, believable crowd motion, no readable fake brand names.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 16:9, 480p, 4 seconds, audio off, generated in 2 min 20 s): A steaming wok stall, string lights and lanterns, and a small group sharing dishes at a wooden table, under warm practical light against a blue evening sky. Did not follow: the camera drifts sideways past one stall rather than walking through the market, and the large handwritten signs carry made-up characters. No brand names are readable, as asked, but check signage before using a clip like this for a real place.

Example 14 test render, the full 4-second clip (looped).

15. Coastal train window

A traveler sits beside a train window as cliffs and blue sea pass outside, sunlight flickers across the face, and the traveler looks up from a notebook; intimate medium close-up, camera fixed inside the carriage, natural reflections, calm cinematic travel diary, consistent face and hands.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 16:9, 480p, 4 seconds, audio off, generated in 2 min 18 s): The traveler writes in a notebook beside the window, then looks up toward the sea and cliffs as sunlight crosses his face and shirt. The camera stays inside the carriage and his face and hands stay consistent. Did not follow: the cliffs outside barely move across the 4 seconds, so the train reads as parked rather than moving.

Example 15 test render, the full 4-second clip (looped).

16. Old-town morning

A baker opens wooden shutters on a quiet cobblestone street while bicycles pass in the distance and sunlight reaches the upper buildings; slow lateral dolly, soft early-morning light, realistic European old-town atmosphere, subtle ambient movement, no crowds or modern billboards.

Try this prompt in SeedVideo AI

17. Tropical waterfall reveal

A narrow jungle path opens onto a tall waterfall surrounded by dense tropical plants; begin close behind broad leaves, move forward into a wide reveal, humid mist catching sunlight, rich natural greens, immersive expedition documentary style, physically realistic water flow.

Try this prompt in SeedVideo AI

18. Desert camp at blue hour

Lanterns illuminate a small desert camp as two travelers pour tea beside a low table and the last blue light fades behind the dunes; gentle circular camera move, warm lantern light against cool sky, quiet luxury-travel mood, realistic fire and fabric motion, no oversized moon.

Try this prompt in SeedVideo AI

Social Media Ad Prompts

When to use them: short vertical ads, launch announcements, app promos, creator hooks, before-and-after stories, and fast product education. Build the first frame around one instantly readable idea.

19. Three-second problem hook

Vertical 9:16 video: a creator stares at an empty editing timeline, taps the desk in frustration, then looks toward a bright finished-video preview appearing beside the laptop; quick push-in, clean home studio, high-contrast natural light, relatable creator-ad style, readable composition with space for headline text.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 9:16, 480p, 4 seconds, audio off, generated in 2 min 14 s): The creator frowns at an editing timeline in a clean home studio, the camera pushes in, and he ends with a small smile at a finished autumn clip. Did not follow: the finished preview appears inside the editing app on the laptop screen, not beside the laptop, and the screen shows a made-up editor interface.

Example 19 test render, the full 4-second clip (looped).

20. Before-and-after room styling

Vertical split-screen: the same small living room begins plain on the left and transforms into a warm, styled space on the right as furniture and lighting change in controlled stages; locked camera, fast but smooth transitions, bright lifestyle-ad lighting, room geometry remains identical.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, text to video, 9:16, 480p, 4 seconds, audio off, generated in 2 min 15 s): One of the two closest matches in the set. The frame splits down the middle, the left half stays plain, and the right half gains a rug and side table, then a lamp, flowers, and warm light, in stages. The camera stays locked and the walls, floor, and window line stay the same. Did not follow: only a small one: the shared sofa grows a different arm on the right side, so it no longer lines up across the split.

Example 20 test render, the full 4-second clip (looped).

21. Restaurant menu teaser

Three fast close-up shots of a chef finishing a signature dish: sauce poured, herbs placed, then the plated meal turned toward camera; vertical framing, rhythmic match cuts, warm restaurant lighting, rich food texture, premium social ad, realistic hands and utensils.

Try this prompt in SeedVideo AI

22. Fitness challenge opener

A runner ties one shoe at dawn, stands, and launches onto an empty riverside path; vertical low-angle sequence, three concise cuts, cool morning light becoming warm, motivational but realistic sports content, natural body motion, clear negative space for challenge text.

Try this prompt in SeedVideo AI

23. Small-business packing story

A small-business owner folds tissue paper around a handmade ceramic cup, adds a thank-you card, closes the box, and smiles toward camera; vertical overhead shot transitioning to eye level, soft window light, authentic behind-the-scenes style, consistent cup design and packaging.

Try this prompt in SeedVideo AI

24. Fast feature demonstration

Vertical screen-side lifestyle shot: a designer selects one image, enters a short video direction, and reacts to the resulting animated concept; quick over-shoulder framing, bright modern workspace, concise educational-ad pacing, avoid showing a detailed fake interface, focus on the human workflow and result.

Try this prompt in SeedVideo AI

Image-to-Video Prompts

When to use them: animating product photos, portraits, illustrations, concept art, real-estate images, or campaign stills. Upload the image first, then describe motion without rewriting its visual identity.

25. Portrait with subtle movement

Preserve the person's face, hairstyle, clothing, and background composition. Add a gentle breeze moving only loose hair and fabric, one natural blink, and a small shift of gaze toward the window. Slow camera push-in, soft unchanged lighting, realistic restrained motion, no face redesign.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, image to video with the photo as first frame, 480p, 4 seconds, audio off, generated in 5 min 2 s): The face, hair, blue shirt, wall, and soft light all match the source photo, and the man shifts his gaze to the side with restrained motion. Did not follow: the push-in is barely visible, and the source photo has no window, so the look goes to an empty side of the frame. We did not check frame by frame for the single blink.

Example 25 test render, the full 4-second clip (looped). The source image, an AI-generated stand-in, is the first frame.

26. Product photo turntable

Keep the product shape, materials, colors, logo, and label exactly consistent with the reference image. Animate a slow 20-degree turn while a soft highlight travels across the surface. Camera remains stable, background stays unchanged, no new accessories or packaging.

Try this prompt in SeedVideo AI

Test result (Seedance 2.5, image to video with the photo as first frame, 480p, 4 seconds, audio off, generated in 6 min 22 s): The bottle shape, frosted glass, white cap, and the "SERUM" label stay as in the source image, a highlight slides across the glass, and the camera and background stay still. Did not follow: the bottle turns much more than 20 degrees, so by the last frame only the edge of the label is visible.

Example 26 test render, the full 4-second clip (looped). The source image, an AI-generated stand-in, is the first frame.

27. Illustrated landscape parallax

Preserve the illustration style and all major shapes. Create layered depth: foreground grass moves slightly, middle-ground trees shift with gentle parallax, clouds drift slowly, and the camera moves forward. No new buildings, characters, or style changes.

Try this prompt in SeedVideo AI

28. Interior design walkthrough

Keep the room layout, furniture, materials, and decor unchanged. Animate a slow stabilized camera move from the doorway toward the window while curtains move lightly and sunlight shifts across the floor. No object morphing, no added furniture, straight architectural lines.

Try this prompt in SeedVideo AI

29. Food photography animation

Preserve the plated dish and table arrangement. Add subtle steam, a slow camera orbit of less than 15 degrees, and a small light change that reveals texture. Keep ingredients, plate shape, garnish, and portion size unchanged; no utensils entering the frame.

Try this prompt in SeedVideo AI

30. Fashion campaign still

Preserve the model's identity, outfit construction, accessories, and pose. Animate a controlled turn of the head, slight fabric movement, and a slow sideways camera slide. Maintain editorial lighting and background color, realistic anatomy, no changes to garment details.

Try this prompt in SeedVideo AI

Camera Movement Prompts

When to use them: when the subject is simple but the shot needs to feel directed. One deliberate move reads better than three competing ones.

31. Slow push-in on a decision

A woman in a gray wool sweater sits at a kitchen table at dawn reading a handwritten letter, then folds it and looks toward the window; slow push-in from a medium shot to a close-up over the full clip, soft window light from the left, muted warm palette, no cuts, face and hair stay consistent, no extra people.

Try this prompt in SeedVideo AI

32. Orbit around a still object

A ceramic teapot on a rough oak table with steam rising from the spout; a smooth 180-degree orbit at eye level, shallow depth of field, morning side light, quiet still-life realism, the teapot never changes shape or pattern, background stays soft and unchanged.

Try this prompt in SeedVideo AI

33. Crane reveal of a landscape

Start low behind tall grass on a hillside, then crane upward to reveal a river valley with a small stone village under overcast light; steady vertical rise, wide lens, natural desaturated greens, slow pacing, no people, no text, no lens flare.

Try this prompt in SeedVideo AI

34. Handheld follow through a doorway

A courier in a plain orange jacket walks through a narrow apartment corridor, opens a door, and steps onto a sunny balcony; restrained handheld tracking from behind, slight natural sway, exposure shifts from dim hallway to bright daylight, realistic footsteps, no cuts, jacket color and bag stay the same.

Try this prompt in SeedVideo AI

35. Whip pan between two subjects

Two chess players face each other across a marble table in a quiet library; a fast whip pan from the older player's hands to the younger player's face as a piece is captured, then a locked close-up on the board, warm lamp light, one whip pan only, no additional camera motion, both faces stay consistent.

Try this prompt in SeedVideo AI

Character and Dialogue Prompts

When to use them: short scenes with one or two speaking characters. Models with native audio, such as Seedance 2.5, can time a single line to an action, so keep dialogue to one or two sentences and say exactly when it lands.

36. One-line apology at a doorstep

A man in his thirties holding a small paper bag stands on a rainy porch and says, "I brought the good coffee this time," then gives a small, hopeful smile; medium shot from inside the doorway, warm interior light against cool rain outside, natural lip sync, quiet rain ambience, no music, one line of dialogue only.

Try this prompt in SeedVideo AI

37. Two friends sharing a secret

Two teenage friends sit on a school gym floor; one leans in and whispers, "Don't tell anyone, but I got in," and the other covers her mouth in surprise; static two-shot at eye level, soft overhead light, subtle echo of a large room, no background extras, faces and clothing consistent throughout.

Try this prompt in SeedVideo AI

38. Silent reaction with ambient sound

An elderly fisherman on a wooden pier reads a message on an old phone, lowers it, and looks out at the gray sea without speaking; slow side push-in, overcast light, gulls and gentle water lapping as the only audio, no dialogue, no music, weathered face stays consistent.

Try this prompt in SeedVideo AI

39. Product line delivered to camera

A barista in a plain black apron holds a paper cup toward the camera and says, "Single origin, no sugar, exactly how you like it," with a quick nod; locked medium close-up in a bright cafe, soft front light, natural speech pace, subtle cafe ambience under the line, no on-screen text or logos.

Try this prompt in SeedVideo AI

40. Argument that ends in laughter

A couple in a small kitchen argues over a burned pan; she says, "That was the last one," he replies, "It was already old," and both start laughing; medium two-shot with a slight handheld feel, warm evening light, realistic overlapping laughter, no background music, one continuous take.

Try this prompt in SeedVideo AI

Vertical Hook Prompts for Short Video

When to use them: 9:16 clips where the first second has to stop a scroll. Lead with the payoff, keep motion readable on a phone, and leave room for captions.

41. Before-and-after in one frame

Vertical 9:16, a cluttered desk splits down the middle: the left half stays messy while the right half tidies itself object by object in a fast stop-motion feel; locked overhead shot, even daylight, playful pacing, no hands visible, the desk and objects remain the same across both halves.

Try this prompt in SeedVideo AI

42. Ingredient drop reveal

Vertical 9:16, fresh strawberries drop one by one into a glass of sparkling water in slow motion, then a fast cut to the full glass with condensation; macro locked shot, bright backlight, crisp bubbles, no hands, no text, the glass shape never changes.

Try this prompt in SeedVideo AI

43. Three-second question hook

Vertical 9:16, a young man looks straight at the camera in a sunlit bedroom and asks, "Would you wear this to a wedding?" then steps back to reveal a bright green suit; static shot, natural window light, natural lip sync, quiet room tone, subject centered with headroom for a caption.

Try this prompt in SeedVideo AI

44. Fast unboxing rhythm

Vertical 9:16, hands open a plain kraft box on a white table and pull out a pair of unbranded white sneakers, tissue paper lifting with each beat; four quick top-down inserts, then a hero shot of the shoes, bright soft light, snappy pacing, no logos, sneakers stay identical in every insert.

Try this prompt in SeedVideo AI

45. Satisfying loop

Vertical 9:16, a single drop of honey falls onto a stack of pancakes and spreads slowly to the edges, ending on a frame that matches the start so the clip loops cleanly; locked macro shot, warm morning light, no hands, no text, pancakes unchanged from first to last frame.

Try this prompt in SeedVideo AI

Reference-to-Video Prompts

When to use them: you already have the identity, product, or motion on file. Give each uploaded asset one job, name it, and tell the model which asset wins if two references disagree. These follow the reference-to-video pattern used by Seedance 2.5, where uploads are addressed as @image1, @video1, and @audio1.

46. Product identity from a photo, motion from a clip

Use @image1 for the bottle's shape, cap, label placement, and color; follow the slow tabletop orbit and final front-facing stop from @video1; place the bottle on a dark stone counter with soft side light; keep the product geometry higher priority than any shape in the motion clip, no extra objects.

Try this prompt in SeedVideo AI

47. Character sheet performs a new action

Use @image1 as the only identity reference for the character: face, hair, jacket, and body proportions; make the character walk across a rooftop at sunset, stop at the edge, and turn toward the camera; steady tracking shot, warm backlight, no second person, do not copy the background from the reference sheet.

Try this prompt in SeedVideo AI

48. Environment plate with a new subject

Use @image1 only for the environment: the empty subway platform, its tiles, lighting, and depth; add a courier in a plain yellow raincoat waiting with a bicycle; medium wide shot, slight camera drift, fluorescent green-white light, ambient station hum, keep the platform layout exactly as shown.

Try this prompt in SeedVideo AI

49. Match a dance move from a reference video

Use @video1 for the motion only: copy the jump, spin, and landing with the same timing; the performer is the character in @image1, wearing the outfit shown there; set the action in a bright dance studio with mirrors, locked wide shot, upbeat but silent, keep hands and feet contact points clean.

Try this prompt in SeedVideo AI

50. Previs to final look

Use @video1 as the shot plan: preserve its camera path, shot order, and character positions; replace the gray placeholder set with the night market environment from @image1 and the placeholder figure with the character from @image2; do not inherit any gray materials or block shapes from the previs.

Try this prompt in SeedVideo AI

Short examples users can adapt for their own AI video projects.

How to Adapt These Prompts in SeedVideo AI

SeedVideo AI provides text-to-video and image-to-video workflows, so the right adaptation depends on whether the model must invent the first frame or preserve one you already have.

1. Choose one shot goal. Define the single action the viewer should understand.
2. Select text-to-video or image-to-video. Use the text-to-video generator for scenes from scratch; use the image-to-video generator when composition or identity already exists.
3. Copy the nearest example. Keep its structure, then replace subject, location, materials, brand details, and mood.
4. Remove unnecessary instructions. Every phrase should control subject, motion, camera, lighting, style, timing, or constraints.
5. Generate a simple draft. Test the core subject motion before adding a complex camera move.
6. Review one failure category at a time. Check identity, geometry, motion, camera, lighting, then background.
7. Revise one variable. If the subject works but the camera does not, change only the camera direction.
8. Create separate shots for separate ideas. Edit several focused clips together instead of asking one prompt for a full commercial.

Practical workflow example

Suppose you want a product launch clip for a watch:

• Draft 1: Keep the camera locked and test the watch rotation.
• Draft 2: Preserve the motion and refine highlights and background.
• Draft 3: Add a restrained macro push-in.
• Final pass: Confirm the dial, markings, case shape, and logo remain consistent.

This staged workflow gives you clearer feedback than rewriting the whole prompt after every result. For multi-shot prompts that need one character to stay consistent across a longer clip, run the final pass on Seedance 2.5 — it holds identity for up to 30 seconds. When a finished clip only needs a few more seconds, continue it from its last frame with the AI video extender instead of regenerating the whole shot. If you are still deciding which generation model fits the job, compare the workflows in Best AI Video Generators in 2026.

Common Prompt Mistakes

FAQ

What is the best format for an AI video prompt?

Use subject + action + environment + camera + lighting + style + constraints. Put the most important instruction early, and keep the prompt focused on one shot.

How long should an AI video prompt be?

Long enough to remove ambiguity, but not so long that several instructions compete. A useful prompt is often two to five sentences. Precision matters more than word count.

Should I include camera movements in every prompt?

No. Use a locked camera when subject motion or product accuracy is the priority. Add a pan, orbit, tracking move, crane, or push-in only when it supports the shot.

How are text-to-video prompts different from image-to-video prompts?

Text-to-video prompts describe the entire first frame and its motion. Image-to-video prompts should preserve the uploaded image and focus on motion, camera behavior, timing, and elements that must not change.

Why do AI video subjects change during a clip?

The prompt may ask for too much motion, combine several actions, or fail to state preservation constraints. Simplify the action, reduce the camera move, and explicitly protect identity, product geometry, or composition.

Can I use these prompts for vertical social videos?

Yes. Add “vertical 9:16 composition,” keep the main subject near the center-safe area, create a readable first frame, and leave negative space if you plan to add captions or a call to action.

Do the same prompts work with every AI video model?

The structure transfers, but models interpret motion, camera language, references, and constraints differently. Keep the creative goal constant, test a short draft, and adjust the wording based on the selected model’s output.

What do I change when I move a prompt from Seedance 2.5 to Veo 3.1 or Kling 3.0?

Cut it to one shot and one action for Veo 3.1's 4 to 8 second clips, remove second-range timing, and write the sound you want because Veo always generates audio. For Kling 3.0, keep one or two beats inside 15 seconds and use the Kling O3 or Motion Control variants when a reference image or video should drive the result. The model table above lists the panel differences.

Can I paste these prompts into Seedance 2.5?

Yes. Every example is written as a single shot brief, which is how Seedance 2.5 reads prompts. For 20 to 30 second clips, add timing beats and an ending frame; the Seedance 2.5 Prompt Guide shows the seven-field version of the same idea.

Do these AI video prompts actually work?

We rendered 10 of them on Seedance 2.5 at 480p and 4 seconds on October 14, 2026. Single-action prompts such as the running shoe and the before-and-after room came back close to the text. The most common misses were a dropped last step, made-up lettering on labels and signs, and a skipped orbit or push-in. Each tested example shows its clip and the miss.

How do I write prompts for 30-second AI videos?

Plan two to four beats with rough timestamps, keep one subject and one location, decide how the clip ends, and state what must not change. Longer clips fail on continuity, not on detail, so spend the extra words on constraints rather than adjectives.

Start Creating

Choose one example, replace the details with your own subject and goal, and generate a short first pass in the text-to-video generator. If you are brand new to the workflow, start with How to Make AI Videos; when you need concepts rather than structures, browse 40 AI Video Ideas. For deeper prompting fundamentals, use the Text to Video Prompt Guide; for reference-driven work, continue with the Image to Video AI Guide.