Playcut v2 is coming MCP · Actor v2 · & much more
Prompt library

AI video prompts sorted by job, each tagged with a model

AI video prompts sorted by job, from product reveals and B-roll to loops, animated stills and extensions, each tagged with a video model Playcut runs.

Updated · Playcut prompt library

40 prompts, eight jobs, one rule: each prompt describes a single shot, with its camera move, subject, action, place, light and sound. The chip on every card names the video model Playcut runs that fits the job, plus the ratio, the length and the credit cost at 720p.

This page is text only, with no sample renders.

How to write AI video prompts that survive the model

Write one shot per prompt, built from six parts. Google's prompting guide for Veo 3.1 and ByteDance's prompt guide for Seedance 2.0 use the same ingredients, and both treat sound as something you write down. Google leads with the camera and ByteDance with the subject, so the cards follow each vendor's order:

  1. Shot and camera: "low tracking shot", "slow push-in".
  2. Subject: concrete nouns, "matte black wireless headphones", not "a sleek product".
  3. Action: one thing that finishes inside the clip.
  4. Setting: place and time of day.
  5. Light and style: direction, palette, film look.
  6. Sound: a quoted line with its speaker, then effects and ambience.

Ask for one camera move per shot; ByteDance's guide says so, and it is a safe default on every model we run. Unsure what a move looks like? Pick a camera move and see what each term means.

Leave readable text and logos for the edit. Alibaba's prompt guide for Wan lists exact text legibility among the things to avoid.

Write exclusions as what you want to see, "an empty beach at dawn" rather than "no people", as Google advises for Veo. HappyHorse 1.1 rejects negative prompts outright.

Which video model reads which prompt

You pick the model for each clip in the video studio's model picker, and nothing switches it for you. Start with Veo 3.1, the default video model, and change it when the job calls for a frame, a price or an input Veo doesn't offer.

Video models Playcut runs, what each is best for, how it names reference images, typical credits at 720p, and its main limit
Model Best for Reference images Typical credits (720p) The catch
Veo 3.1 Veo 3.1 prompts Spoken lines and written sound; all five modes No tag: describe the subject. Up to 3, 16:9 only 1,280 for 8 s 16:9 and 9:16 only; Google caps prompts at 1,024 tokens
Seedance 2.0 Seedance 2.0 prompts 480p drafts, 4k finals, 21:9 frames "the watch from Image 1". Up to 9 (Playcut's API docs write @Image1) 105 for 5 s (50 at 480p) Rejects any input image with a face in it
HappyHorse 1.1 HappyHorse 1.1 prompts Nine ratios, including 4:5 and 9:21; speaks quoted lines [Image 1] with a description. Up to 9 280 for 5 s Rejects negative prompts; cuts text past 5,000 characters
Wan 2.7 Wan 2.7 prompts A photo animated to your own audio track; clip edits Image 1, no brackets. Up to 5 200 for 5 s Alibaba rewrites and expands the prompt
Grok Imagine Grok Imagine prompts Cheap text-to-video drafts up to 15 s Up to 7; xAI now documents a different tag format, so no reference cards here 120 for 6 s 720p at most
Grok Imagine 1.5 Animating a still None 192 for 6 s Image to video only, 720p at most

Veo 3.1 is the only model we run that covers text, image, reference, first and last frame, and extension; it can't edit an uploaded clip. It also costs the most per second, so draft on Seedance 2.0 or Grok Imagine and save Veo for the takes that need its sound.

Product reveals

A reveal is one camera move that ends on the product. Seedance 2.0 can hold a real product from a photo, and HappyHorse 1.1 covers frames Veo 3.1 can't, such as 4:5.

  1. Headphones lit by a moving spotlight

    Veo 3.1 16:9 · 8 s · 1,280 cr
    Slow push-in on matte black wireless headphones resting on a folded grey wool blanket. A warm spotlight sweeps across the ear cups until the whole pair is lit. Shallow depth of field, charcoal background. SFX: a soft fabric rustle, then a low synth swell. Ambient noise: quiet room tone.

  2. Wristwatch from your product photo

    Seedance 2.0 1:1 · 5 s · 105 cr
    The mechanical wristwatch from Image 1 lies on a slate tray. The camera makes one slow orbit as morning light slides across the dial and the second hand ticks. Close-up, rich detail, cinematic texture. Audio: crisp ticking and faint birdsong outside. Keep it subtitle-free, no added text, no watermark.

    Attach a product-only photo; Seedance 2.0 rejects faces.

  3. Perfume atomizer in a 4:5 frame

    HappyHorse 1.1 4:5 · 5 s · 280 cr
    A frosted glass perfume atomizer stands on wet black stone while a fine mist drifts past it. The camera cranes down slowly from above the cap to eye level, catching a thin line of backlight along the glass. Deep amber and black palette, soft haze, calm and luxurious.

  4. E-scooter unfolds on a wet street

    Veo 3.1 9:16 · 6 s · 960 cr
    Low tracking shot along a wet city pavement at blue hour. A folded grey e-scooter unfolds with a click, its stem locking upright and its headlight blinking on. Shop lights reflect in the puddles. SFX: a metallic click, then a soft electric hum. Ambient noise: distant traffic and light rain.

  5. Two-shot watch caseback reveal

    Seedance 2.0 16:9 · 8 s · 168 cr
    Shot 1: extreme close-up of a watch's exhibition caseback, the balance wheel beating behind the glass, fixed camera. Shot 2: the camera pulls back slowly as a hand turns the watch face-up on a leather pad. Warm tungsten light, HD, rich details. Audio: fast ticking. Keep it subtitle-free, no watermark.

Explainers and announcements to camera

Put the words in quotes and say who speaks them. Veo 3.1 and HappyHorse 1.1 speak a quoted line; Seedance 2.0 lip-sync is best-effort, so its card has the shortest line.

  1. Warehouse safety briefing

    Veo 3.1 16:9 · 8 s · 1,280 cr
    Medium shot, a shift supervisor in a hi-vis vest stands in a bright warehouse aisle between tall pallet racks, holding a tablet. She looks into the camera and says, "Forklifts own the yellow lanes. People walk on the green." Ambient noise: a distant reversing beeper and the hum of ceiling fans.

  2. Release note read-out

    HappyHorse 1.1 16:9 · 8 s · 448 cr
    A product manager in a navy sweater stands at a desk in a sunny loft office, a laptop open beside him, and speaks straight to camera, saying: "New this week: reports can land in your inbox every Monday at nine." Steady medium close-up, soft daylight, calm and friendly.

  3. Conference recap from the foyer

    Veo 3.1 9:16 · 8 s · 1,280 cr
    Handheld medium shot in a busy conference foyer after the last session. A host in a green blazer turns to the camera and says, "Three days, forty talks, one takeaway: ship smaller, ship sooner." People pack up behind her. Ambient noise: crowd chatter and a rolling suitcase.

  4. Museum audio-guide welcome

    Veo 3.1 16:9 · 8 s · 1,280 cr
    Slow dolly through an empty museum gallery at opening time, sunlight falling across marble busts and a long wooden bench. A calm older narrator's voice says, "Welcome. Take your time in this room; everything here was made in one summer." SFX: soft echoing footsteps. Ambient noise: a large, quiet hall.

  5. Pollen forecast at the map wall

    Seedance 2.0 16:9 · 6 s · 126 cr
    A presenter in a yellow blazer stands before a large wall screen showing a map of green and orange zones and gestures toward the orange patch as it pulses. Medium shot, fixed camera, bright studio light. She says: "Pollen peaks after lunch." Keep it subtitle-free, no watermark.

Cutaway B-roll

B-roll carries sound more than story. Veo 3.1 and Seedance 2.0 always generate audio, so describe the noise you want. For inserts where the sound doesn't matter, Grok Imagine is the cheapest video model we run at 720p.

  1. Subway doors close

    Veo 3.1 16:9 · 6 s · 960 cr
    Close-up of subway train doors sliding shut on a crowded platform, then the carriage pulls away and the camera holds on the empty yellow line. Cool fluorescent light, slight film grain. SFX: a two-tone door chime and a pneumatic hiss. Ambient noise: the rumble of the departing train.

  2. Wind across a wheat field

    Seedance 2.0 21:9 · 5 s · 105 cr
    Wide shot, slow lateral tracking along a wheat field at golden hour as wind moves across the stalks in long waves, a line of poplars on the horizon. Warm tones, cinematic texture. Audio: wind through dry grass and a skylark overhead. Keep it subtitle-free, no watermark.

  3. Keyboard under a desk lamp

    Grok Imagine 16:9 · 6 s · 120 cr
    Overhead shot of hands typing on a mechanical keyboard in a dim office, a desk lamp throwing a warm pool of light, a mug of tea steaming beside the keyboard. The camera drifts slowly to the right.

  4. Wave on black sand

    Veo 3.1 9:16 · 4 s · 640 cr
    Low angle at the waterline on a black volcanic sand beach as a wave rushes in, spreads thin and pulls back, leaving white lace of foam. Overcast light, cool blue-grey palette. SFX: the wave's crash and the fizz of retreating foam. Ambient noise: steady wind.

  5. Steam from a street at dawn

    Seedance 2.0 480p · 5 s · 50 cr
    Dawn on an empty avenue, fixed camera: steam rises from a manhole cover while a street-sweeper truck passes in the background, orange lights turning. Audio: a low hiss and the truck's distant brushes. Keep it subtitle-free, no watermark.

    Set the frame to 16:9.

  6. Sticky notes on a glass wall

    Grok Imagine 16:9 · 5 s · 100 cr
    Slow push-in on a glass meeting-room wall covered in rows of yellow and pink sticky notes, bright daylight. A breeze from an opening door lifts the corners of the notes, and one flutters to the floor.

Loops for social feeds

A loop needs its last frame to match its first. Say so in the prompt, keep the motion cyclical and the camera locked. For the cleanest join, use a first and last frame mode on Veo 3.1 or Seedance 2.0 and attach the same still to both slots.

  1. Lava lamp, same frame in and out

    Veo 3.1 9:16 · 8 s · 1,280 cr
    A lava lamp on a windowsill at night: amber wax blobs rise, split and sink back so the scene returns exactly to the starting frame. Locked-off camera, soft city lights beyond the glass. SFX: a faint bubbling. Ambient noise: a quiet room.

    Attach the same image as the first and last frame. First-and-last-frame clips on Veo 3.1 are always 8 seconds.

  2. Newton's cradle on a walnut desk

    Seedance 2.0 1:1 · 6 s · 126 cr
    A chrome Newton's cradle on a walnut desk: the end balls swing in a steady rhythm and the motion ends exactly where it began. Fixed camera, soft window light. Audio: a crisp click on every strike. Keep it subtitle-free, no watermark.

    Use Seedance's transition mode with one face-free still as both the first and the last frame.

  3. Cake stand, one full turn

    HappyHorse 1.1 4:5 · 6 s · 336 cr
    Seamless loop: a slice of layered sponge cake on a slowly turning cake stand makes exactly one full rotation, strawberries and cream catching soft window light against a pastel pink background. Static camera. The final frame matches the first frame.

  4. Paper boat in a puddle

    Wan 2.7 1:1 · 5 s · 200 cr
    A seamless looping shot of a paper boat rocking gently on a round puddle while ripples spread outward and fade, the same slow rhythm repeating. Top-down view, fixed camera, soft overcast light, muted blue and grey palette.

    Wan 2.7 prompts are expanded before rendering, so state the loop plainly.

  5. Candle flame on black

    Grok Imagine 9:16 · 6 s · 120 cr
    A seamless loop of a single candle flame flickering in a dark room, the wax glistening and a thin ribbon of smoke curling up and fading. Locked-off close-up, warm amber on black.

Animate a still you already made

The still already sets the subject, place and style, so the prompt only needs motion and camera, which is how Alibaba's own guide frames image to video. Veo 3.1 letterboxes a still that isn't 16:9 or 9:16; Wan 2.7 and HappyHorse 1.1 follow the still's shape.

  1. Portrait takes off his glasses

    Grok Imagine 1.5 16:9 · 6 s · 192 cr
    The man in the image takes off his reading glasses, folds them and looks up at the camera with a small nod. The camera holds still at eye level.

    Attach the still: Grok Imagine 1.5 is image to video only.

  2. Illustrated city map comes alive

    Veo 3.1 16:9 · 8 s · 1,280 cr
    Small trams glide along the drawn routes of the illustrated city map, clouds drift across the paper, and the camera slowly pushes in toward the old town. SFX: a soft tram bell. Ambient noise: faint street bustle.

  3. Soup on a rainy day

    Wan 2.7 5 s · 720p · 200 cr
    Steam curls up from the bowl of tomato soup, a hand dips a slice of toast into it, and the camera tilts slowly down from the rain-streaked window to the table.

  4. Isometric office wakes up

    HappyHorse 1.1 5 s · 720p · 280 cr
    The isometric office illustration comes alive: tiny figures walk between the desks, a printer slides out a page, the wall clock's hands turn, and the camera orbits a quarter turn to the right.

    Leave the ratio on its default; HappyHorse 1.1 rejects any other setting on image to video.

  5. Wet dog shakes off

    Grok Imagine 1.5 16:9 · 8 s · 256 cr
    The dog in the photo shakes water off its coat in slow motion, droplets flying, then trots toward the camera. The camera stays low at the dog's eye level.

Lock a look with reference images

Each model names attached images its own way, so never cross-paste a tag. Veo 3.1 takes no tag: describe the subject. HappyHorse 1.1 reads [Image 1], Wan 2.7 reads Image 1, and Seedance 2.0 gets the subject plus "from Image 1", as ByteDance writes it.

  1. One character, described not tagged

    Veo 3.1 16:9 · 8 s · 1,280 cr
    The woman in the mustard wool coat and round glasses walks along a canal bridge at dusk, stops, and checks a paper map. The camera tracks beside her at shoulder height. Soft lamplight, light mist over the water. Ambient noise: footsteps on stone and a distant church bell.

    Veo 3.1 reference mode is 16:9 and 8 seconds only, with up to three images.

  2. Lamp and reading nook

    HappyHorse 1.1 16:9 · 5 s · 280 cr
    The ceramic table lamp from [Image 1] switches on in the reading nook from [Image 2]. The camera drifts in slowly as warm light spreads over the armchair and the bookshelves. Evening, calm, interior magazine look.

  3. Balcony with a named character

    Wan 2.7 16:9 · 5 s · 200 cr
    The man in the corduroy jacket from Image 1 waters the plants on the balcony from Image 2, then leans on the railing and looks out over the rooftops. Medium shot, slow push-in, late afternoon sun.

  4. Product on a set, both from photos

    Seedance 2.0 4:3 · 5 s · 105 cr
    The perfume atomizer from Image 1 stands on the marble ledge from Image 2 as petals drift down around it. The camera makes one slow push-in. Cinematic texture, rich details. Audio: a faint breeze and a soft chime. Keep it subtitle-free, no added text, no watermark.

    Product and set photos only: faces are rejected, AI-generated ones included.

  5. Two hikers from two photos

    HappyHorse 1.1 16:9 · 6 s · 336 cr
    The woman in the orange shell jacket from [Image 1] and the man in the grey beanie from [Image 2] reach a cairn on a ridge, high-five, and look out over the valley. The camera cranes up slowly behind them. Crisp morning light.

Extend or re-cut a clip you like

When a clip is nearly right, change it instead of starting over. Veo 3.1 extends its own 720p clips by 7 seconds a call. Wan 2.7 and HappyHorse 1.1 edit a clip you upload, and both bill the source seconds plus the output seconds.

  1. Runner reaches the end of the pier

    Veo 3.1 +7 s · 1,280 cr
    The runner slows at the end of the pier, stops at the railing and catches her breath as the sun clears the horizon. The camera circles slowly around to face her. SFX: her breathing and gulls overhead. Ambient noise: waves under the pier.

    Extension works only on 720p clips Veo generated, within Google's two-day storage window. A voice carries on only if someone speaks in the source's last second.

  2. Drone drops the parcel

    Veo 3.1 +7 s · 1,280 cr
    Continuing the shot, the delivery drone lowers the parcel onto the doorstep, releases it with a click and climbs back into the sky. The camera stays fixed on the front door. SFX: rotor whine fading upward. Ambient noise: suburban birdsong.

    Each extension call is billed like an 8-second 720p clip and adds 7 seconds, up to 20 times.

  3. New view out of the office window

    Wan 2.7 Edit · 5 s in · 400 cr
    Change the view through the office windows from a car park to a sunlit harbour with sailboats, and keep the people, the desks and the camera move exactly as they are.

    Wan 2.7 edits a 2 to 10 second MP4 or MOV. A 5-second source with a 5-second output bills 10 seconds at 720p.

  4. Swap the mug in her hand

    HappyHorse 1.1 Edit · 6 s in · 672 cr
    Replace the plain white mug in the presenter's hand with the orange enamel travel mug from [Image 1], keeping her grip, the lighting and everything else in the clip the same.

    HappyHorse 1.1 uses the first 15 seconds of a 3 to 60 second source and returns the same length, so a 6-second clip bills 12 seconds.

Cheap drafts before the final render

Block the idea where a take is cheap, then spend credits on the final. Seedance 2.0 at 480p is the cheapest per second of the video models we run, and Grok Imagine is the cheapest at 720p.

  1. Rooftop skateboard, three shots

    Seedance 2.0 480p · 10 s · 100 cr
    Shot 1: wide shot of a skateboarder rolling across an empty rooftop car park at dusk. Shot 2: close-up of the wheels crossing painted lines. Shot 3: the skater kicks the board up and walks off. Audio: wheel roll and a board clack. Keep it subtitle-free, no watermark.

    Once the shot order works, re-run the same prompt at 1080p (600 credits for 10 seconds).

  2. Paper plane through an office

    Grok Imagine 16:9 · 6 s · 120 cr
    A paper plane glides the length of an open-plan office, dips under a hanging lamp and lands on a desk beside a laptop. The camera follows it at head height.

    If the path reads well, add SFX and Ambient noise lines and move it to Veo 3.1 for the final (960 credits for 6 seconds).

  3. Scramble crossing from above

    Seedance 2.0 480p · 8 s · 80 cr
    Top-down view of a busy city intersection at noon: pedestrians cross from all four corners as the lights change, then cars move again. Fixed camera, timelapse feel. Audio: crossing-signal beeps and traffic. Keep it subtitle-free, no watermark.

    For the final, 1080p costs 60 credits a second; 4k costs 108 if it has to hold up on a big screen.

  4. Van in a rainy village

    Grok Imagine 16:9 · 8 s · 160 cr
    A small delivery van drives slowly down a narrow village street of pastel houses in the rain, headlights on, and stops outside a bookshop. The camera tracks alongside at window height.

    Keep the timing, then re-run on Veo 3.1 at 1080p with rain and tyre sounds written in (1,280 credits).

  5. Carpenter at the bench

    Seedance 2.0 480p · 6 s · 60 cr
    Medium shot, slow push-in: a carpenter runs a hand plane along an oak board, long pale shavings curling off the blade and dropping to the floor. Warm workshop light. Audio: the rasp of the plane and shavings rustling underfoot. Keep it subtitle-free, no watermark.

    A person is fine here because this is text to video. The 720p final costs 21 credits a second.

From copied card to finished clip in Playcut

Copy a card, open the Playcut video studio, and then:

  1. Pick the model named on the card in the model picker. Veo 3.1 is selected by default.
  2. Set the ratio and length from the card's chip, and 480p where the chip says so.
  3. Attach any images the card names, in the order it numbers them.
  4. Paste and generate. Credits are held when you submit and refunded if the job fails or you cancel it.

The prompt box takes up to 8,000 characters, far more than any card here needs.

To see every model side by side, compare every model we run in credits, or read how the Playcut AI video generator handles text, image and reference modes. When a clip misses, fix a video prompt symptom by symptom.

Need the still first? Start from AI image prompts grouped by job, then animate the result with a card from the group above. Ad formats with a creator on camera live on the UGC ad prompts page.

Each model also has its own prompt page:

FAQ

What makes a good AI video prompt?

One shot, described in order: the framing and camera move, a concrete subject, one action that finishes inside the clip, the setting, the light, and the sound. Concrete nouns beat adjectives, and a spoken line belongs in quotes with a named speaker.

How long should an AI video prompt be?

One short paragraph per shot is plenty; the prompts here run from about 30 to 50 words. Playcut's prompt box takes up to 8,000 characters, but Google caps Veo 3.1 at 1,024 tokens and HappyHorse 1.1 silently cuts anything past 5,000 characters.

Can AI video models render text or logos?

Poorly, and it gets worse in motion. Alibaba's Wan prompt guide lists exact text legibility among the things to avoid, and ByteDance suggests telling Seedance to keep a clip subtitle-free. Add titles, captions and logos in the edit.

Which AI video model should I use for a given job?

Use Veo 3.1 when someone speaks or the sound matters, HappyHorse 1.1 for 4:5 and other unusual frames, and Wan 2.7 to animate a photo to audio you recorded. Use Seedance 2.0 for cheap 480p drafts or 4k finals with no faces in the inputs, and Grok Imagine for cheap drafts.

How many camera moves fit in one AI video clip?

One per shot. ByteDance's Seedance 2.0 guide asks for a single camera movement per shot, and it is a safe default on the other models too. For a second move, write a second shot or extend the clip.