प्रोडक्शन गाइड

AI video prompt library: 20 copyable prompts by scenario

Prompt advice ages faster than models, but prompt structure does not. Every prompt in this library follows the same skeleton — one shot per prompt, camera anchored, motion as concrete clauses, mood stated once — because that is what survives model updates. Copy them, swap the nouns for yours, and skim the notes on why each line is there. For model-specific structure, pair this with the Kling and Seedance prompt guides.

अपडेट

केस साक्ष्य: मूल ब्रेकडाउन रिपोर्ट पढ़ें · केस फ़्रेम और शॉट समय देखें

प्रकाशित टूल रिपोर्ट पर आधारित। निर्माण वर्गीकरण विश्लेषणात्मक निर्णय हैं; सुझाए चरण मूल प्रोजेक्ट या सत्यापित रीमेक नहीं हैं।

Talking photo prompts

Talking photos live or die on the face, so the prompt spends its words there and leaves voice to the audio track. “She looks into the camera, blinks naturally, a slight head sway as she speaks, a warm smile at the final word. Static medium close-up, soft window light from the left.” “He begins serious, one slow eyebrow raise mid-sentence, lands on a grin. Locked-off close-up, even indoor lighting.” “The grandmother glances down at the photo in her hands, then up to camera, eyes bright. Slow push-in from medium to close.” The pattern: identity of the gaze first, then one or two motion events, then camera, then light. Never more than one expression change per generation — stack them and the jaw artifacts.

Product ad prompts

Ads sell exactness, so the prompt buys atmosphere and leaves the product to footage or a first frame. “Slow orbit around the perfume bottle, dust motes drifting through a warm shaft of light, deep amber background. Camera arc, 15 degrees per beat.” “The sneaker hovers, rotating slowly, city dusk reflected in the gloss beneath it. Rising crane shot.” “Steam rises from the mug on a wooden counter, morning light crawling across the grain, hands enter frame and wrap around the ceramic. Locked-off medium.” Note what is absent: brand names, label text, prices — generation gets those wrong, so the prompt never asks.

Holiday and seasonal prompts

Holiday spots want warmth over realism, and prompts should lean into it. “Snow falls softly on a lit village street, one family walks away from camera toward the glow, breath visible. Slow tracking follow.” “Golden light strings sway in a dark room, a cat's tail enters frame and bats one bulb. Static wide, shallow depth.” “Autumn leaves spiral across a park bench, one leaf lands on the empty seat and stays. Locked-off medium.” The rule of one: one season, one light source, one subject, one motion. Holiday scenes collapse into greeting-card mush exactly when prompts stack three of each.

Music video beat prompts

Music videos are generated in beats, one prompt per beat, cut on the rhythm. “Dancer mid-leap frozen against a graffiti wall, dust hanging in the air, hard side light. Snap zoom in.” “Singer turns from the window to the lens, one tear, city bokeh behind. Slow 90-degree arc.” “Crowd hands rise in silhouette against strobing stage light, confetti falls in slow motion. Low angle, static.” Write beats on the beat sheet before prompting: prompt 1 = verse mood, prompt 2 = pre-chorus lift, prompt 3 = hook. Videos that feel generated are usually videos prompted as one long shot instead of eight beats.

Atmosphere and b-roll prompts

B-roll is the safest thing to generate and the easiest to overwrite. “Rain streaks a train window, countryside blurring green and grey. Locked-off close-up, droplets in focus.” “Waves fold onto black volcanic sand at dawn, foam drawing back. Slow pull-back from macro to wide.” “Ceiling fan turns above an empty diner counter, dust in the light, a door chime sways. Static medium.” B-roll prompts win by being boring: one surface, one weather, one motion, no subject. The subject is what the cut gives it.

Avatar spokesperson prompts

Avatar pipelines take direction differently: personality in the prompt, delivery in the voice script. “A warm, credible woman in her forties, smart-casual, standing in a bright office. Professional but human; occasional hand gestures.” “A young presenter, energetic, denim jacket, plain studio backdrop. Speaks quickly, smiles between points.” Keep avatar prompts about casting — age, warmth, energy, wardrobe — and let the script carry pacing. The classic mistake is writing delivery directions (“pauses meaningfully”) into the prompt where the TTS never reads them.

अक्सर पूछे जाने वाले प्रश्न

How long should an AI video prompt be?

Shorter than you think. One shot, one anchored camera, one or two concrete motion events, one line of mood. A prompt that runs past four clauses is usually trying to be two shots, and the model will pick the wrong half. If you need more, cut the video, not the prompt.

Do these prompts work on Kling 3.0, Seedance 2.5 and Runway Gen-4.5?

The structure does, on all of them: these models differ in what they render best, not in how they read prompts. Swap the nouns, keep the skeleton — camera anchor first or last, motion as concrete clauses, mood once. For version-specific habits, read the which-AI-video-model guide alongside this library.

Should I write negative prompts for AI video?

Mostly no. Current video models weigh positive description far more than exclusion lists, and a negative line (“no blur”) often just injects the word into attention. Fix problems by changing the positive instruction (“razor-sharp focus on the eyes”) or by re-cutting the shot.

Can I reuse one prompt across a whole video?

Reuse the skeleton, never the prompt. A video reads as generated when every shot shares the same camera and cadence. Vary one axis per shot — camera, distance, or motion — and keep the mood constant; that is what the music-video beat pattern in this library does.

पुन: प्रयोग योग्य विश्लेषण टेम्पलेट

TALKING PHOTO
1. looks into camera, natural blinks, slight head sway, smile at end. Static MCU, soft window light.
2. serious → one eyebrow raise → grin. Locked-off CU, even light.
3. glances down at photo, up to camera, eyes bright. Slow push-in M→CU.
PRODUCT
4. slow orbit, dust motes in warm light shaft, amber bg. Camera arc.
5. product hovers rotating, dusk reflected beneath. Rising crane.
6. steam rises, light crawls across wood, hands wrap the mug. Locked-off M.
HOLIDAY
7. snow on lit street, family walks toward glow, breath visible. Tracking follow.
8. light strings sway in dark room, cat tail bats a bulb. Static wide, shallow DOF.
9. leaves spiral over a bench, one lands and stays. Locked-off M.
MUSIC BEATS
10. dancer mid-leap, dust hanging, hard side light. Snap zoom.
11. singer turns to lens, one tear, city bokeh. Slow 90° arc.
12. crowd hands in silhouette, strobe, confetti slow-mo. Low angle static.
B-ROLL
13. rain on train window, countryside blurs. Locked-off CU.
14. waves on black sand at dawn, foam recedes. Macro→wide pull-back.
15. ceiling fan over empty counter, dust in light. Static M.
AVATAR
16. warm credible woman 40s, smart-casual, bright office. Human, gestures.
17. energetic young presenter, denim jacket, plain studio. Quick, smiles.
REMEMBER: one shot, anchored camera, concrete motion, mood once.

असली रिपोर्ट के साथ अभ्यास

See the skeleton in action: the music-video case’s per-shot prompts follow exactly this structure — beat, camera, motion, mood — across 118 shots. Read three of them next to their evidence frames.

केस खोलें और रिपोर्ट डाउनलोड करें →

अपना वर्कफ़्लो जारी रखें