The first three seconds decide everything. That's the window you have to stop a scroll, trigger a re-watch, or earn a share. The best AI video prompts are built around that window — not around what's visually interesting at the 10-second mark.

This list collects 30 prompts that consistently produce hooks strong enough to survive the opening seconds. They're organized by format, so you can match the prompt type to the platform and audience you're targeting.

What Separates a Good AI Prompt from a Great One

Most prompt guides focus on aesthetics: lighting, camera angle, color grade. Those matter. But the prompts that produce viral-worthy clips share three specific structural traits:

1. Action is front-loaded. The most interesting moment happens immediately, not after a cinematic build. Prompts should describe what's happening before they describe how it looks.

2. The unexpected element is concrete. "Something surprising" is too vague for any model. "A businessman in a suit running from a wave of rubber ducks" is concrete enough to render.

3. One subject, one action. Complex multi-subject prompts produce cluttered frames at short durations. The viewer's eye needs one clear place to land.

30 Best AI Prompts for Viral Videos

Format 1: Transformation (Stops Scrolls Fast)

Transformation clips work because the before and after create a natural loop — viewers re-watch to see the transition again.

A bustling New York street freezes mid-motion. Every person, car, and pigeon turns to solid gold statuary. Morning light.
A wilted, dying rose in a glass vase. The water turns golden, the rose blooms in reverse time-lapse.
A stone wall cracks apart and living ivy grows through every gap in seconds.
A person's silhouette dissolves into a flock of birds that scatter across a sunset sky.
A handwritten letter crumbles and the words lift off the page, forming a flying flock of paper cranes.

Format 2: Surreal Humor (Most Shareable)

Humor drives shares. Surreal humor — where the world is photorealistic but the scenario is absurd — works especially well because it feels like a strange dream clip rather than an obvious joke.

A golden retriever in a business suit presents a PowerPoint slide to a room of attentive cats.
A group of penguins in chef hats run a sushi restaurant. One penguin is the maître d'.
Serious Olympic swimmers competing in a pool filled with lime Jell-O. Full crowd, slow-motion cameras.
A tiny medieval village inside a snow globe. Knights fight a dragon. Snow falls outside the glass.
An otter in a three-piece suit argues with a lobster in a conference room. Both have lawyers.

Format 3: Cinematic Atmosphere (High-Save Content)

These prompts produce visually striking clips that users save for inspiration. Save rate is an underrated virality signal on Instagram and TikTok.

A foggy mountain road. A lone cyclist appears from the mist, rides past the camera, dissolves back into fog. Slow shutter.
Cherry blossoms fall in slow motion over a quiet Tokyo alley at dusk. Warm amber street lights.
A lighthouse beam sweeps across a rough Atlantic sea at night. The beam catches a passing whale.
A child releases a paper boat into a flooded street. The boat sails away between glowing windows.
Rain falls in reverse over a city skyline. Droplets rise from puddles and return to storm clouds.

Format 4: "What If" Reality (High Re-Watch)

"What if" clips tap into an innate curiosity loop — viewers watch again to understand what they just saw.

What if ants were the size of horses? An ant colony commutes through a suburban neighborhood in rush hour.
What if clouds fell to the ground? A city street covered in cloud material. Pedestrians wade through wisps.
What if the ocean was transparent? A glass-clear sea showing the seafloor 30 meters below a sailing boat.
What if shadows had their own shadows? A figure walks, four layered shadows fan out behind them.
What if gravity switched off for exactly five seconds? A park scene. People, leaves, and a dog float then drop.

Format 5: One-Second Hook Openers (Platform-Optimised)

These prompts are specifically designed for the first-second hook — ideal for TikTok's auto-scroll feed where the opening frame has to earn attention before audio even kicks in.

EXTREME CLOSE-UP: An eye opens. Its iris is an entire galaxy, stars rotating slowly.
A city building collapses in slow motion — but in reverse, reassembling itself brick by brick.
A hand opens flat. A miniature thunderstorm swirls in the palm. Lightning strikes the fingers.
A blank white canvas. Paint explodes from the center in a perfect radial burst — then resolves into a landscape.
A person snaps their fingers. The room behind them instantly rearranges into a different decade.

Format 6: Satisfying Loops (Re-Watch Optimised)

Loop-friendly clips inflate watch time metrics. These prompts describe scenes that end where they begin.

Molten gold pours over a dark surface and cools into a perfect sphere. The sphere melts back to liquid.
A dandelion puff scatters in slow motion. Seeds float back and reassemble the flower.
An hourglass empties as the camera slowly rotates 180°. It finishes upright, ready to begin again.
A city day-night cycle compressed to 10 seconds. Lights flick on in buildings as the sun sets. Dawn returns.
Smoke rises from a candle. The smoke reverses, descending back into the flame. The flame extinguishes and relights.

Matching Prompts to Models

Not all AI video models handle these formats equally:

Format	Best Model	Why
Transformation	Kling 3 / Seedance 2	Handles physics-defying motion well
Surreal humor	Kling 3	Maintains coherent subjects through absurd scenarios
Cinematic atmosphere	Veo 3.1 / Sora 2	Superior lighting and environmental detail
"What if" reality	Sora 2	Strong world-physics simulation
One-second hooks	Any	Model matters less than prompt specificity
Satisfying loops	Veo 3.1	Consistent frame-to-frame quality for smooth loops

aicut gives you access to all these models — Sora 2, Veo 3.1, Kling 3, and Seedance 2 — from one dashboard, so you can test the same prompt across models and pick the best output without switching between platforms.

Even the best base prompt needs iteration. Here's the fastest path to a usable output:

Step 1 — Strip to the core. Remove adjectives until you have: Subject + Action + Setting. Generate once.

Step 2 — Add specificity. Add the one detail that makes it unexpected (e.g., "all attendees are rubber ducks"). Generate again.

Step 3 — Add camera and light direction. "Slow motion", "macro lens", "golden hour" lock in the mood. Final generation.

Three generations to a polished clip is realistic. Don't over-engineer the first draft.

Platform-Specific Tweaks

TikTok: Favor prompts #6–10 (humor) and #21–25 (one-second hooks). Caption the clip with a "wait for it" text overlay to pump up watch time.

Instagram Reels: Prompts #11–15 (cinematic atmosphere) perform well on Explore because aesthetic quality drives saves. No text overlay needed — let the visuals breathe.

YouTube Shorts: Transformation (#1–5) and loop prompts (#26–30) work best. Shorts viewers tolerate a slightly slower build than TikTok. Add a spoken or text hook in the first second.

Key Takeaways

Front-load the action. The most interesting moment belongs in the first 1–2 seconds.
The six most reliable formats are: transformation, surreal humor, cinematic atmosphere, "what if" reality, one-second hooks, and satisfying loops.
Match your prompt format to the model's strength — Veo 3.1 and Sora 2 for realism, Kling 3 for dynamic action, Seedance 2 for iterative testing.
Iterate in three steps: core → unexpected detail → camera/light direction.

Try these prompts now on aicut — all major AI video models in one place, no separate subscriptions needed.