10 AI Video Prompts for Reels & TikTok

Short-form is where AI video earns its keep. A 4-minute AI short looks impressive but rarely converts; a 7-second scroll-stopper built from a good prompt can pull thousands of views a day. These AI video prompts for Reels and TikTok are written for the formats that actually perform on short-form in 2026: instant hooks, product motion, seamless transitions, and loopable clips — all native vertical 9:16.

How to use them: copy the prompt block, paste it into your generator (Runway, Pika, Luma, or a script-to-video tool like InVideo or Pictory), then swap the bracketed placeholders for your subject, brand colors, and product. Generate everything in 9:16 vertical — some tools default to 16:9 and will letterbox your output, which kills the feed look. Add your own captions and trending audio after generation; AI video supplies the visuals, your edit supplies the virality. For prompts that lean photorealistic rather than stylized, see [AI Video Prompts That Look Real]](/ai-video-prompts-that-look-real/), and for a full tool rundown, [Best AI Video Generators 2026]](/best-ai-video-generators-2026/).

1. The 3-Second Hook

The first three seconds decide everything. This prompt builds a high-energy opener with aggressive motion so the viewer's thumb stops.

Extreme close-up of [subject: e.g. a dripping honey dipper over toast], macro detail, droplets falling in slow motion, then camera whips back fast to reveal the full scene. High contrast, saturated colors, cinematic shallow depth of field. Vertical 9:16. Motion: fast push-out after the whip, subtle handheld shake. No text, no watermarks.

Best on image-to-video tools (Runway, Pika, Luma) where you supply the macro still and let the model animate it. Tweak the "whip back" speed — too slow and the hook drags, too fast and viewers can't parse the reveal. Common failure: the slow-motion section renders with smeared, melty detail. Fix: generate the close-up as a still first, then animate at lower motion intensity and trim the first second.

2. Product Reveal Spin

A floating 360-degree product spin is the workhorse of AI product content — clean, reusable, and impossible to shoot without a turntable.

Studio product shot: [product: e.g. matte black serum bottle with gold cap] floating in a softly lit infinite-white studio void, rotating slowly 360 degrees, gentle rim light tracing its edges, tiny dust particles drifting in the air. Seamless loop motion. Vertical 9:16, photorealistic, crisp reflections, no hands, no text.

Suits text-to-video generators with strong product coherence (Luma, Runway). The tweak that matters most: name the material precisely ("matte black glass, brushed gold cap") — vague materials give you plastic-looking renders. Common failure: the product label or logo warps mid-rotation. Fix: keep the spin slow (one full turn per 4–5 seconds) and generate at least three variants, picking the frame-stable one.

3. The Seamless Transition

Transitions carry "satisfying" and transformation content. This prompt chains two scenes through one continuous camera move, which reads as a single slick edit.

One continuous shot: [scene A: cluttered desk] fills the frame, camera pushes forward through a burst of paper particles, emerging into [scene B: the same desk perfectly organized, warm morning light]. Same camera angle and desk position in both scenes for a match cut. Smooth fast dolly, photorealistic, vertical 9:16, no text.

Works on tools with start/end-frame control — feed scene A as the first frame and scene B as the last where supported (Runway supports this; Luma handles it via keyframes). Tweak the particle density: too many particles and the "emerging" moment gets lost. Common failure: the desk's position jumps between scenes, breaking the match cut. Fix: lock the description of the desk's position ("center frame, same oak desk") in both halves of the prompt verbatim.

4. Before/After Split Screen

The classic transformation frame, generated in one pass so both halves share lighting and style — no compositing needed.

Split screen, vertical divider down the middle: left half shows [before state: dull, dimly lit small bedroom, grey bedding, clutter], right half shows [after state: same bedroom, bright warm light, white and sage bedding, tidy, plant in corner]. Camera slowly pushes in on both halves equally. Identical camera height and angle on both sides. Photorealistic interior photography style, vertical 9:16, no text.

Suits text-to-video models with strong spatial control (Runway, Pika). Tweak the "same bedroom" anchor words — repeat the room's fixed elements ("same window on the back wall") so the model keeps geometry consistent. Common failure: the divider drifts or one half renders a completely different room. Fix: regenerate with the room description identical in both halves and add "mirrored composition" to the prompt.

5. Kinetic Text Overlay Style

For talking-head-free explainers, this generates the kinetic background that your captions sit on top of — bold, punchy, and endlessly reusable.

Abstract kinetic background: [brand colors: deep navy and electric cyan] liquid gradient waves flowing upward, subtle geometric grid lines pulsing to an implied beat, occasional soft light flares. Dark, premium, high-end motion-graphics feel. Center of frame kept intentionally empty and dark for text overlay. Seamless loop, vertical 9:16, no text, no logos.

Best on motion-graphics-friendly generators and loop-capable tools (Runway, Pika), or as a background layer under an OpusClip-style caption edit. Tweak the "empty center" instruction — if you skip it, the generator fills the middle with shapes that fight your captions. Common failure: visible seams when looping. Fix: request "seamless loop" explicitly and, if the tool allows, set the clip length to a short 3–4 seconds, which loops more cleanly than long generations.

6. UGC-Style Testimonial Framing

AI "customers" are a compliance minefield — never present them as real people — but the framing style works for founder-led and demo content where you voice the script yourself.

Handheld-style medium shot of [person: e.g. a woman in her 30s, casual denim jacket] in a bright modern kitchen, holding [product] up toward camera at arm's length, natural window light, slight handheld wobble, shallow depth of field, authentic smartphone-footage aesthetic. She is not speaking — neutral pleasant expression, subtle head nod. Vertical 9:16, photorealistic, no text overlays.

Use image-to-video (Runway, Luma) with a real photo or a clearly-labeled AI avatar from HeyGen or Colossyan as the base, then record your own voiceover. Tweak the "not speaking" line — lip-sync on AI faces still looks wrong in 2026; a nodding, non-speaking clip under your VO reads far more real. Common failure: the uncanny dead-eye stare. Fix: keep the shot 3–5 seconds max and cut before the viewer studies the face; add your captions and B-roll quickly.

7. The Perfect Loop

Loopable clips get rewatched, and rewatches feed the algorithm. This prompt builds a visual that returns exactly to its start.

[subject: e.g. coffee being poured into a glass cup] in an endless repeating cycle: the pour stream never empties, steam rises and curls back down into the cup, camera orbits slowly around the cup. Motion designed so the last frame matches the first frame exactly. Warm cozy lighting, macro detail, vertical 9:16, seamless infinite loop, no text.

Suits Pika and Runway, both of which handle loop-oriented motion well. Tweak the orbit speed — a full orbit in 5–6 seconds feels hypnotic; faster feels dizzy. Common failure: the "last frame matches first" instruction gets ignored and the loop visibly jumps. Fix: generate a 4-second clip, then in your editor duplicate it with a 0.5-second crossfade at the seam — the crossfade hides what the model couldn't close.

8. Trending Meme Format

Meme formats burn out in weeks, so this is a template, not a one-off: swap the bracketed format for whatever's trending and generate fast.

[trending format: e.g. "expectation vs reality" — left-to-right camera pan]: first a glamorous influencer-style setup of [subject], then camera pans right past a motion-blur wipe into the chaotic real version: [messy reality version of the same subject]. Fast comedic timing, punchy, exaggerated expressions if people appear, vertical 9:16, bright flat lighting like phone footage, no text.

Best on fast, cheap generators (Pika, Luma) where you can iterate daily without burning credits. Tweak the wipe element — "motion-blur wipe" gives the model a concrete transition to render instead of an awkward morph. Common failure: the two halves bleed together into one muddled scene. Fix: name the pan direction explicitly and describe the wipe as a full-frame element ("the wipe fills the entire frame for 3 frames").

9. Cinematic B-Roll

Atmospheric B-roll is the filler that makes short-form feel premium — generate a small library of these and reuse them across videos.

Slow cinematic drift over [scene: rain-streaked city window at night, neon signs blurred outside], anamorphic lens flare, shallow focus shifting from the raindrops to the bokeh beyond, moody teal-and-orange grade, film grain, subtle camera float. No people, no text. Vertical 9:16, photorealistic, 24fps film feel.

Suits Runway and Luma at their highest quality settings — this is where you spend the good credits, since one great B-roll clip gets reused for months. Tweak the focus-shift direction: "foreground to background" reads cinematic; the reverse reads like a mistake. Common failure: the grade comes out flat and video-gamey. Fix: name the grade ("moody teal-and-orange, crushed blacks") and add "film grain" — grain alone sells realism more than any other single word.

10. CTA End Card

Every short needs an ending that asks for the follow. Generate the card; add your handle and the follow button animation in your editor.

Clean motion-graphics end card: [brand color] gradient background, soft animated light sweep moving left to right, subtle floating particles, large empty center space for text, gentle pulsing glow at the bottom third where a button would sit. Premium, minimal, 3 seconds, vertical 9:16, no text, no logos, seamless hold on last frame.

Best on any generator with clean graphic output (Runway, Pika), then finish in VEED or Descript where you add your actual handle, caption, and CTA text. Tweak the "empty center" and "no text" lines — AI-rendered text still garbles, so always add real text in your editor. Common failure: the model sneaks in gibberish text despite the instruction. Fix: add "absolutely no letters, no words, no typography" — redundant phrasing works better than a single "no text".

Adapting These Prompts

Three levers turn a decent prompt into a reliable one. First, aspect ratio: always state 9:16 vertical in the prompt itself, and double-check your tool's project settings — a 16:9 generation cropped to vertical loses half its composition. Second, negative prompts: where your tool supports them, add "blurry, watermark, text, logo, extra fingers, deformed hands" — hands remain AI video's weakest point in 2026, so keep them out of frame or out of focus whenever the prompt allows. Third, iterate in batches: never judge a prompt on one generation. Run three variants, keep the best 20%, and save the winning prompt text with the settings you used — over a month that file becomes your personal prompt library, worth more than any listicle.

For scripted or talking-head style shorts, Syllaby can turn an idea into a captioned vertical video in one pass, and Synthesia covers the avatar-presenter format if you want a consistent on-screen face across a series.