Home/Blog/2026年AI動画プロンプト集:コピペできる22例
ガイド10 分で読める

2026年AI動画プロンプト集:コピペできる22例

ダンス・映画風ストーリー・カメラワークとライティング・製品・クリエイター風広告・写真から動画へ・Sora 2風ショットリスト、タスク別22例のAI動画プロンプト。本番recipe抜粋とテンプレート、制約句付き。

執筆 ZNIX Team · AI動画リサーチ・モデル評価
公開日

A good AI video prompt is a shoot instruction, not a pile of adjectives: lock identity, order the motion beats, fix the camera, and write the constraints. This library collects 22 copy-ready prompts — two excerpted from real production recipes behind our tools, the rest clearly labeled editorial templates — organized by dance performance, cinematic story, camera and lighting, product ecommerce, creator-style ads, photo-to-video and a Sora 2 style section, plus a checklist of constraint clauses that prevent the usual failures.

Anatomy of a video prompt

Five slots: (1) subject and identity lock — which reference maps to whom and which features stay; (2) motion beats — ordered by seconds, not "make it move"; (3) camera and frame — static front-facing, 9:16, one continuous shot; (4) constraint clauses — negatives like no cuts / no text / no extra limbs; (5) audio declaration — silent or original score. Missing any slot invites the model to improvise, and improvisation is where clips break.

Dance & performance prompts

Production recipe excerptHotel Lobby AI →
Create a new Hotel Lobby-inspired performance in a seamless burnt-orange studio booth with one silver microphone hanging centrally from the ceiling. Use image 1 for the LEFT subject and image 2 for the RIGHT subject. Keep these two identities distinct and preserve each face, hair, fur and outfit from their own reference. Two best friends playfully trade rap gestures, with relaxed head nods and a shared groove. Static front-facing camera, vertical 9:16 composition, both subjects fully visible with room for hands or paws. One continuous shot, natural motion, consistent identities and clothing, no extra subjects, no cuts, no text, no logos, no face morphing or extra limbs. Generate a silent clip with no music or voices.
Editorial templateRaindance AI →
A single dancer performs an original rain-themed groove on a wet neon-lit street at night, reflections rippling under each step. Medium full-body framing, static camera at chest height, vertical 9:16. Motion beats: 0-2s settle into the groove, 2-5s two arm waves and a spin, 5-8s freeze pose with rain streaks catching the light. Preserve the face and outfit from the reference photo. One continuous shot, no cuts, no text, no logos, no extra limbs, no face morphing. Silent clip.
Editorial templateSeedance 2.0 →
Photo-to-dance: animate the person in the reference photo performing a short original choreography — two step-touches, one shoulder roll, one turn — in a plain studio with soft key light. Keep the face, hairstyle and clothing exactly as in the reference. Full-body framing with headroom, static camera, 9:16, one continuous take. Natural joint motion, no extra people, no text or watermark, no face morphing. Generate silent video.

Cinematic story prompts

Production recipe excerptAI Zombie Video →
Create an original 15-second cinematic zombie love story using both reference images. Image 1 is the adult human survivor: healthy natural skin and normal eyes throughout; preserve their face, hair and clothing. Four ordered story beats: 0-4s, inside a cold blue-gray abandoned cabin, open with a clear close-up of the zombie face; 4-7s, the survivor lowers an unfired prop handgun to the floor and opens their arms; 7-11s, the loved one approaches slowly and they embrace gently; 11-15s, hard match cut on the same embrace pose to a warm golden meadow at sunset, both healthy, happy and alive. Maintain identity, relative size, outfits and screen positions across shots. Natural hands and limbs, no extra characters, no text, logos or watermark. No blood, gore, open wounds, gunfire or aggressive violence.
Editorial templateAI Video Generator →
A 10-second cinematic memory piece: open on a close-up of hands holding an old photograph, slow push-in; match cut on the same hand position to the same hands young again in a sunlit kitchen; hold the warm frame for three seconds. Shallow depth of field, film grain, gentle camera drift only. Preserve the two identities from the references. No text, no logos, no extra characters, no cuts other than the single match cut. Original ambient score, no dialogue.
Editorial templateAI Short Video →
Vertical 9:16 micro-story in three beats: 0-3s a commuter misses the train and sighs; 3-6s a stranger hands them a coffee and smiles; 6-9s both laugh as the next train arrives, warm morning light. Static camera per beat, one cut per beat, consistent faces and outfits across beats, natural hands, no text or watermark, no extra crowd members appearing mid-shot. Silent.

Camera & lighting prompts

Camera and light are the two controllable variables behind “cinematic”: write the camera as one continuous move, and the light as direction plus quality plus change. These are editorial templates — the camera language reads on Kling, Seedance and Wan; validate complex moves at short duration first.

Editorial templateAI Video Generator →
One continuous environment take: the camera glides forward through a night market at walking pace, passing lantern stalls on both sides, gentle handheld sway, warm tungsten pools against cool blue shadows. No cuts, no teleporting, no new subjects appearing mid-frame, one unbroken eight-second move, 16:9. Silent.
Editorial templateAI Video Generator →
Unified cinematic brief: a lone traveler crosses a rain-slick platform at dusk. Subject: dark wool coat, identity locked from the reference. Motion: three ordered beats — walk, pause, look up. Camera: slow lateral dolly right to left, 35mm look, shallow depth of field. Light: single sodium lamp key, wet reflections, light fog. Constraints: one continuous shot, no text, no extra characters, no face morphing. Six seconds, 16:9, silent.
Editorial templateAI Video Generator →
Lighting fade effect: the scene starts in cold blue pre-dawn light and slowly warms to golden sunrise over eight seconds, the transition continuous and even, shadows rotating naturally with the light source. Static camera on the same skyline, no cuts, no flicker, no exposure jumps, no text. Silent.
Editorial templateAI Video Generator →
Practical-light mood study: a desk lamp clicks on inside a dark room, warm cone of light reveals dust motes, the rest of the frame stays in deep shadow. Slow push-in, high contrast, film grain, one continuous shot, five seconds, no flicker, no text, no extra objects. Silent.

Product & ecommerce prompts

Editorial templateAI Product Video →
Clean ecommerce turntable: the product from the reference image rotates 360 degrees on a matte white pedestal, softbox lighting from both sides, subtle contact shadow. Static camera at product height, 1:1 frame, 6 seconds, one continuous shot. Preserve label text, logo placement and color exactly; no distortion of packaging, no extra objects, no hands, no background changes. Silent.
Editorial templateAI Product Video →
Hand-held demo: two hands enter frame, pick up the product from the reference, turn it once to show the back label, then set it down on a linen tablecloth. Natural window light, shallow depth of field, 9:16, 8 seconds. Keep packaging graphics legible and unchanged; natural finger anatomy, no extra hands, no text overlays, no jump cuts. Silent.
Editorial templateAI Ads Video →
Lifestyle insert: the product from the reference sits on a kitchen counter while steam rises from a mug beside it; slow lateral dolly left to right, morning light through a window, 16:9, 5 seconds. Product label stays sharp and unwarped; background bokeh only; no people, no text, no logos added, no camera shake. Silent.

Creator-style ad prompts

Ground rule: a synthetic persona has no first-hand experience, so never frame it as a real customer testimonial. Use these templates for hook testing, demos and explainers, and label synthetic content per platform rules.

Editorial templateAI UGC Video →
Creator-style hook demo, vertical 9:16: a presenter holds the product from the reference at chest height, looks into the lens and gestures once toward the label, then tilts the product to catch the light. Handheld selfie framing with mild natural shake, room ambient light, 6 seconds. Preserve the presenter face from the reference; natural hands, no text overlays, no logos added, no jump cuts. Original voiceover-free silent clip for later captioning.
Editorial templateAI UGC Video →
Problem-solution explainer in two beats: 0-4s the presenter shrugs beside a messy desk; 4-8s the same desk is tidy with the product from the reference centered, presenter nods once. Static selfie camera, consistent face and outfit across the cut, 9:16. No fabricated before/after claims in visuals, no text, no extra people, natural hands. Silent.
Editorial templateAI Ads Video →
Three-variant hook test base: same presenter, same room, same product reference; variant A opens with the product close to lens, variant B opens with a raised eyebrow and a point at the label, variant C opens with the product already in use on the desk. 9:16, 5 seconds each, static handheld framing, consistent identity and lighting across variants, no text, no logos, no cuts inside a variant. Silent.

Photo-to-video prompts

Editorial templateAI Video Generator →
Living portrait: animate the reference photo with a slow 3D parallax push-in, subtle hair and fabric movement from a light breeze, eyes blink once naturally. Keep the face geometry, expression and background exactly as in the photo; no head turn, no smile change, no new objects. 4 seconds, 9:16, one continuous shot, no text or watermark. Silent.
Editorial templateExplore Gallery →
Pet motion from a single photo: the pet from the reference tilts its head once, ears twitch, then settles back into the original pose. Preserve species, fur markings, collar and background; natural animal anatomy, no human-like expressions, no extra animals. 4 seconds, original framing, one continuous shot, no text or logos. Silent.

Sora 2 style prompts

Sora 2’s strength is character identity across cuts and long-form narrative coherence, so its prompt style reads like natural-language shot lists: describe each shot’s scene, action and transition instead of piling on style words. Access caveat: Sora 2 lives inside ChatGPT (Plus ~5 renders/day, Pro ~50/day) and the Sora app only — no developer API — and the original Sora was shut down on April 26, 2026. All cards below are editorial templates (the first is the sample from our Sora 2 model profile); the same shot-list style runs on Hailuo 2.3 / Seedance 2.0 here, billed per credit with no daily quota. Native audio is a Sora 2 Pro feature; keep the silent declaration when running elsewhere.

Editorial templateSora 2 Model Profile →
A young woman walks through a Tokyo alley at night, neon reflections in puddles, then turns and smiles at the camera. 12 seconds.
Editorial templateAI Short Video →
A three-shot narrative short with one consistent character: Shot 1 (0-4s), a barista in a green apron wipes the counter of a quiet cafe at dawn, static medium shot; Shot 2 (4-8s), the same barista hands a paper cup to a customer at the door, over-the-shoulder angle; Shot 3 (8-12s), the customer walks into the morning street holding the cup, slow pull-back. Keep the same face, apron and cafe interior across all shots, natural hands, no extra characters appearing mid-shot, no text or logos. 12 seconds total, 16:9. Silent.
Editorial templateAI Video Generator →
A single continuous scene with declared audio: a street musician plays an acoustic guitar under a bridge while light rain falls, close-up on the strings then a slow tilt up to the face. Audio: solo acoustic guitar and soft rain ambience only, no dialogue, no crowd noise. One continuous 10-second shot, shallow depth of field, warm practical light against cool shadows, no cuts, no text, no extra people. (Audio generation requires Sora 2 Pro; on other models render silent and add sound in post.)
Editorial templateAI Video Generator →
A physics-spectacle hook clip in the viral Sora style: a glass sculpture of a koi fish shatters in ultra slow motion on a black marble table, shards catching a single spotlight, one continuous 8-second shot, macro framing, high contrast, no cuts, no text, no logos, no people. Silent.

Constraint clauses that prevent failures

These clauses are excerpted from our production recipes; pick them by failure mode: identity drift → "preserve each face, hair and outfit from their own reference / keep the two identities separate"; extra limbs → "natural hands and limbs, no extra limbs"; random cuts → "one continuous shot, no cuts"; stray text → "no text, logos or watermark"; face swaps → "no face morphing"; audio runaway → "generate a silent clip with no music or voices". Write constraints as one closing block, not scattered adjectives.

Finally: prompts are free to copy, but generation is billed per each tool’s live credit quote; outputs are probabilistic, so review identity, hands and constraint compliance before publishing. More real outputs with per-item prompts live in the Explore gallery.

よくある質問

What makes a good AI video prompt?
Five slots: identity lock per reference image, motion beats ordered by seconds, camera and frame (static / 9:16 / one continuous shot), a closing block of negative constraints (no cuts, no text, no extra limbs, no face morphing), and an audio declaration. Missing slots let the model improvise, which is where clips break.
Are these prompts free to use?
Yes, copying prompts is free. Generation is billed per tool at the live credit quote shown before rendering; registered users can claim free signup credits subject to daily quotas, and the $3 trial pack covers higher-quality renders.
Do the prompts guarantee the result?
No. Video models are probabilistic: identity, hands and constraint compliance vary by input and model. Review every clip before publishing and regenerate with tightened constraints when a clause is violated.
Which prompts come from real production?
The Hotel Lobby and zombie-story cards are excerpts of the actual recipes behind those tools, labeled as production excerpts. All other cards are editorial templates written for this library and labeled as such; none claims measured benchmark results.
Do these Sora 2 prompts work without a Sora subscription?
The Sora 2 style cards are natural-language shot lists, so the same prompts run on credit-based models like Hailuo 2.3 and Seedance 2.0 on ZNIX with no daily quota. Sora 2 itself is only available inside ChatGPT Plus/Pro or the Sora app (no API), and its native audio generation is a Pro-tier feature — keep the silent declaration when rendering on other models.

さらに読む

執筆者について
ZNIX Team — AI動画リサーチ・モデル評価

ZNIX編集チームは、プラットフォームに搭載された各動画モデルを実際に検証し、ベンダーの宣伝ページではなく生成ログに基づいて執筆しています。

AI動画を作る準備はできましたか?

登録で50無料クレジット — クレジットカード不要。

無料で始める →