A useful AI video prompt says who or what is in the shot, what changes over time, how the camera moves, and what the viewer hears. Here are ten full examples you can copy and adapt for Seedance, Kling and Veo.
Examples 9 and 10 include clips we generated ourselves. The other examples are copy-ready starting points, not a promise that every model will return the same result. All the prompt text is original to OpenMicroDrama and reusable under CC BY 4.0.
1. Seedance: vertical establishing shot
Use this when you need one clear opening image and one camera move.
Vertical establishing shot of a glass skyscraper at night in heavy rain. Only the top floor is lit, warm gold, while the rest of the tower is dark. The camera starts low on the wet street, where taxi lights smear across the puddles, then tilts up slowly along the glass facade to the penthouse windows, where a single silhouette in a suit stands looking down. Cool blue city light against a warm interior glow, light haze, glossy romance-drama look. Sound: <steady rain on pavement>, <distant traffic>, (low, tense piano). No subtitles, no on-screen text, no logos.
Settings: Seedance 2.0 Fast, text-to-video, 9:16, 6s, 720p, audio on.
The slow tilt is the only camera move. Angle brackets mark sound effects and parentheses mark music, so the two do not blur together.
2. Kling: close-up with one small action
Use a tiny facial movement instead of asking for several emotions and actions in one shot.
Extreme close-up, 9:16 vertical. A 32-year-old CEO with a sharp jaw, slicked-back black hair, a charcoal three-piece suit, silver cufflinks and cold dark eyes. He sits at the head of a glass boardroom table and slowly lifts his gaze from a document to look just past the camera, jaw tightening, one eyebrow rising a fraction. Night city lights blur behind him through floor-to-ceiling windows. Shallow depth of field, cool blue key light with a warm rim on his cheek, quiet tension. Audio: low air-conditioning hum, a pen clicking once. No dialogue.
Negative prompt: smiling, talking, extra people, text, subtitles, warped hands, face morphing, blur.
Settings: Kling 3.0 Pro (1080p), text-to-video, 9:16, 5s, audio on.
The order is subject, movement, scene, then camera and light. "No dialogue" keeps the model from inventing a line.
3. Seedance: three-character reference setup
Give every reference image one job, and state the headcount before the shots begin.
@Image 1 is Claire's face and @Image 4 is her full look: sleek platinum bob, black satin dress, red lipstick. @Image 2 is Grant's face, a 35-year-old man in a navy suit with a smug half-smile. @Image 3 is Vanessa's face, a 26-year-old woman in a pink tweed set with an oversized diamond ring. @Image 5 is the living room of the family mansion. Exactly three people in frame, no duplicates.
Shot 1: Over-the-shoulder from behind Claire (@Image 1), looking across a glass coffee table at Grant (@Image 2) and Vanessa (@Image 3) on a white sofa. Claire drops a thin envelope on the table.
Shot 2: Reverse over-the-shoulder from behind Grant. Claire stays standing. She says in English {Open it. Your lawyer already has a copy.}
Shot 3: Close-up on Vanessa as she reads. Her ring hand starts to tremble and she pulls it into her lap.
Cool daylight, polished marble, tense quiet. Sound: <envelope landing on glass>, <paper unfolding>. No background music, no subtitles.
Settings: Seedance 2.0, reference-to-video, 9:16, 10s, 720p, audio on; references: three headshots, one full-body image and one location.
Each lead has a named reference, the location has its own image, and "exactly three people" makes cloning less likely.
4. Seedance: image-to-video reaction shot
Let the start frame carry identity and composition. Use the prompt for motion and sound.
Close-up of Elena, a 24-year-old executive assistant with chestnut hair in a low ponytail, a cream silk blouse and a thrifted tan trench coat, standing in a marble lobby. She looks up from her phone as elevator doors open off-screen. Her polite expression freezes: lips part, she swallows once, and her eyes widen and stay fixed on someone just past the camera. The camera pushes in slowly from her shoulders to her eyes. Soft key light from the left, cool rim light, shallow depth of field. Sound: <elevator chime>, <her breath catching>. No dialogue, no subtitles.
Settings: Seedance 2.5, image-to-video, 720×1280 start frame, 6s, 720p, audio on.
The video follows the start frame's shape, so make that frame vertical before you begin. A swallow and a fixed stare are clearer directions than a vague "shocked".
5. Kling: dialogue across five shots
Label every shot and its duration, and keep dialogue inside the first ten seconds.
@Element1 is Adrian, the CEO. @Element2 is Elena, his assistant. A glass penthouse office at night, rain on the windows.
Shot 1 (3s): Over-the-shoulder from behind @Element1, looking down at @Element2 seated across a black desk. He slides a contract toward her with two fingers.
Shot 2 (3s): Close-up of @Element2, eyes scanning the page, breath catching. She says quietly: "Six months as your fiancée? That's insane."
Shot 3 (3s): Over-the-shoulder from behind @Element2 onto @Element1. He leans back, unbothered, and says: "It's a contract, not a proposal."
Shot 4 (3s): Insert close-up of her hand hovering a pen over the signature line, trembling slightly.
Shot 5 (3s): Close-up of @Element1 watching her, the corner of his mouth almost moving. The rain gets louder.
Cool blue light, warm desk lamp, shallow depth of field. No background characters, no music, consistent room tone.
Settings: Kling 3.0 Omni Pro (1080p), reference-to-video with two character elements, 9:16, 15s, audio on, custom multi-shot (5 × 3s).
Each character has one line in a separate shot. On fal, each Shot line goes in its own multi-prompt entry.
6. Veo: four cuts inside one 8-second clip
Repeat the character description before the timestamped cuts so identity and lighting stay consistent.
Vertical cinematic romance, glossy night interior, the same two characters in every cut. ADRIAN: a 32-year-old CEO with a sharp jaw, slicked-back black hair, cold dark eyes, a charcoal three-piece suit and silver cufflinks. ELENA: a 24-year-old woman with chestnut hair in a low ponytail, a cream silk blouse and a thrifted tan trench coat.
[00:00-00:02] Medium shot: Elena steps out of an elevator into a crowded rooftop party, clutching her coat closed, eyes scanning the room.
[00:02-00:04] Medium close-up: Adrian turns away from a group of investors, sees her, and sets down his glass.
[00:04-00:06] Two-shot: he crosses to her, takes her left hand and lifts it so a diamond ring catches the light. He says to the room: "Meet my fiancée."
[00:06-00:08] Close-up on Elena's face, frozen in shock, as a camera flash pops off-screen.
SFX: party chatter dropping to silence, one glass clinking, a camera shutter. Ambient noise: soft lounge jazz.
Negative prompt: subtitles, captions, on-screen text, watermark, extra rings, warped hands.
Settings: 9:16, 8s, 1080p, audio on, Veo 3.1.
The four cuts have a simple progression: entrance, recognition, announcement, reaction. Only one person speaks.
7. Kling: product advertisement
For a product shot, describe one object, one lighting setup and one camera move. This is an original ad brief, not a rendered case study.
A matte-white cylindrical smart speaker on a warm stone desk at dawn. The camera begins with a macro view of the fabric texture, then pulls back slowly as the indicator ring glows once. Soft window shadows move across the wall. Keep the speaker shape and desk layout unchanged. Quiet premium commercial look, shallow depth of field, warm neutral colors, natural room ambience and a restrained activation chime. No people, no extra products, no text or logos.
Negative prompt: hands, extra speakers, changed logo, warped geometry, subtitles, captions, flicker, abrupt camera movement.
Settings: Kling 3.0 Standard (720p), text-to-video, 9:16, 6s, no dialogue.
The product stays still while the camera and light provide the motion. There is no feature claim to verify and no invented brand name to render.
8. Veo: UGC-style spoken ad
For UGC, keep one speaker, one location and one short line. This is also an original ad brief, not a rendered case study.
Handheld vertical UGC video in a quiet home office. A woman in her late 20s sits at a wooden desk in a plain cream sweater, holding a small white smart speaker beside her laptop. She looks into the lens and says conversationally, "This is the little speaker I keep moving from my desk to the kitchen." She places it down and smiles. Natural morning light, slight handheld movement, realistic room tone, no background music.
Negative prompt: subtitles, captions, on-screen text, extra people, beauty filter, warped hands, product logos.
Settings: 9:16, 6s, 720p, audio on, Veo 3.1 Fast.
There is one speaker and one line, so the model can focus on voice and expression instead of a multi-person dialogue.
9. Seedance: a lie and its giveaway, with our preview
This example uses a specific gesture to reveal what the character is hiding. The clip below is our own Seedance 2.0 render from this prompt.
Close-up of Eli, a soft-spoken 31-year-old man with wire-rimmed glasses and a gray cardigan, across a metal interrogation table under a single overhead light. He keeps his voice gentle. He says in English {I've never been to that lighthouse.} As he finishes the line, his thumb rubs the edge of his glasses frame twice and his eyes flick to the left, then back. The camera holds perfectly still. Cold fluorescent light, gray walls, shallow focus. Sound: <fluorescent buzz>, <chair creak>. No music, no subtitles.
Settings: Seedance 2.0, text-to-video, 9:16, 8s, 1080p, audio on. Our preview was generated at 720p.
A vague "he looks nervous" is harder to act than rubbing the glasses twice and flicking the eyes left. The small gesture is the story.
10. Seedance: small object insert, with our preview
Hands and props are useful when you do not want to keep a face consistent. This clip is also our own render from the exact prompt.
Extreme close-up insert on a dark mahogany desk at night. A man's hand with a silver cufflink slides a thick contract across the wood and taps the signature line twice with a black fountain pen. Rain streaks on the window reflect across the glossy desk. Locked camera, shallow depth of field, focus on the pen tip. Sound: <paper sliding on wood>, <two sharp pen taps>, <rain on glass>. No readable text on the page, no subtitles.
Settings: Seedance 2.0 Mini, text-to-video, 9:16, 5s, 720p, audio on.
There is no face, one hand and one action. The prompt also asks for no readable text; add the actual contract wording in your editor instead of asking the model to draw it.
Common prompt repairs
| Problem | Change this first |
|---|---|
| Unwanted music | Put "no BGM, no music, no score" near the start and repeat it at the end. Add your own music in the edit. |
| Duplicate people | State the exact headcount, name each reference and remove background characters. Split a crowded scene into separate shots. |
| Garbled text | Ask for no readable text, subtitles or logos. Add titles and signs in the editor. |
| Changing faces or outfits | Reuse the same character references and repeat identity details word for word. Do not change the outfit description between shots. |
| Lip sync on the wrong person | Give each speaker a separate shot or a separate generation. Keep lines short and early in the clip. |
| Cropped vertical video | Make the start frame 9:16 before using image-to-video. Do not crop a wide image after generating it. |
Reusable six-line prompt template
Use this as a starting point, then adapt the field names to the model. Seedance, Kling and Veo do not share one perfect syntax.
[Image1 is the character; Image2 is the location.]
Subject: [name, age, hair, outfit]
Action: [one action]
Scene: [place, weather, background people]
Style: [photoreal, light, depth of field]
Camera: [shot size, one camera move], 9:16
Sound: [ambient noise, one line of dialogue]. No subtitles.
If you want a form instead of a blank template, use the AI video prompt generator. Browse the complete Seedance, Kling and Veo libraries for more examples, or plan the shots first with the vertical storyboard template.
Sources
- docs.byteplus.com/en/docs/ModelArk/2222480
- docs.byteplus.com/en/docs/ModelArk/2607689
- fal.ai/models/fal-ai/kling-video/v3/pro/image-to-video
- fal.ai/models/bytedance/seedance-2.5/reference-to-video
- kling.ai/quickstart/text-to-video-prompt-guide
- docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/video-gen-prompt-guide