AI video prompts
Start with this one and change a noun at a time: [Shot 1] 4s: Medium shot, handheld, following @Mara as she crosses a rain-wet market at dusk, lanterns lit behind her. She stops at a stall and picks up a small brass key. Ambient: rain on awnings, distant voices.
The formula
Write each prompt as one shot, in the order a viewer notices things:
| Part | What goes in it | Example |
|---|---|---|
| Framing | One framing: wide, medium, close-up | Medium shot |
| Camera | Still, handheld, tracking, a quick push-in: one movement | handheld, following her |
| Who and where | The person, the place and where things are | @Mara in a rain-wet market at dusk, lanterns behind her |
| The one thing that happens | One action, not three | She picks up a small brass key |
| Sound | Speech and ambience, positively worded | Ambient: rain on awnings |
| Seconds | Enough time for what happens | 4s |
Everything else is the look, and the look is set once for the whole project by the style presets, not repeated in every prompt. How to write one shot · Choosing the look.
Five cinematic shot prompts
| Job | Prompt |
|---|---|
| Opening | Wide shot, high angle, moving forward over a harbour town at dawn; boats below, mist on the water. Ambient: gulls and a distant engine. |
| Tension | Close-up on @Jun's hands as they hover over the red button, fixed camera, a bead of sweat sliding down his wrist. Ambient: a faint electrical hum. |
| Chase | Handheld, following @Mara as she cuts between market stalls, the crowd blurring past, wet stone underfoot. Ambient: shouting, running footsteps. |
| Reveal | Pulling back from @Elena at the window to show the empty ballroom behind her, chairs stacked, dust in a single beam of light. |
| Quiet moment | Medium close-up, fixed camera. @Teo reads the letter by the stove, lips moving slightly; the kettle begins to whistle. No dialogue. |
Notice what is missing: no cinematic, 8K or masterpiece. Those words add nothing the model can place in the frame, push the useful words down the line, and are sometimes painted literally. Why tags do nothing · Cinematic camera moves, in plain words.
Three image prompts for a character
A character sheet is a full-length casting photo on a plain white background, so write the person, not the mood: face first, then build, then one outfit. Give the face one anchor nobody else has.
| Character | Prompt |
|---|---|
| A detective | A woman in her late forties, long face, heavy straight brows, deep-set grey eyes, a small scar through the left eyebrow, grey streak at the temple; tall and spare; a charcoal wool coat. |
| A young baker | A man of about twenty-two, round face, wide-set brown eyes, freckles across the nose, a chipped front tooth, short curly black hair; solid build, slightly stooped; a flour-dusted apron over a green shirt. |
| A fantasy elder | An old woman, narrow face, hooded dark eyes, deep lines at the mouth, long white hair in a single braid, a mole under the right eye; small and upright; a patched blue travelling cloak. |
Words like beautiful or charismatic describe nothing that can be drawn and pull toward a stock face. Describing a face · AI character sheet generator.
Two place prompts
| Place | Prompt |
|---|---|
| An interior | A narrow kitchen in a flat above a shop: a gas stove on the left wall, a small table with two chairs under the window, a door to the stairs at the far end; late afternoon light from the window. |
| An exterior | A market square at dusk, wet cobbles reflecting string lights; stalls with striped awnings along the left side, a stone fountain in the centre, a church tower at the far end. |
Say where the door is, which side the window is on and where the light comes from; a moving camera exposes vague rooms. Your scene image becomes the set.
Two ad prompts
An ad takes a photo and one line rather than a shot list: what it is and who it is for.
| Product | One line |
|---|---|
| A bottle | A recycled-glass water bottle with a bamboo lid, for people who walk to work and hate buying plastic. |
| A shop | Our family bakery on the corner of Elm Street, open since 1987; for people who want bread that is still warm; walk-ins welcome. |
When the line and the photo disagree, the line wins. Writing the one-line brief · AI ad video generator.
Two dialogue prompts
| Moment | Prompt |
|---|---|
| A threat | Close-up on @Brook, jaw set. @Brook (low, through his teeth): You had no right. (voice: Brook) Ambient: a clock ticking in the hall. |
| A confession | Medium shot, fixed camera, @Lina on the balcony at night. @Lina (quiet, not turning): I already knew. (voice: Lina) Ambient: far traffic, a door closing. |
About three English words a second: a short line needs about four seconds with its reaction. Writing dialogue in the storyboard.
Weak prompts and their rewrites
| Weak | Why it fails | Stronger |
|---|---|---|
| A cinematic 8K video of a girl in a city | Tags the model cannot place; no framing, action or place | Medium shot, @Lina walking through a crowded night market, neon signs reflected in the wet ground. Ambient: sizzling woks. |
| No crowd, no rain, no text | Every noun named is a candidate for the picture | An empty street, dry pavement, a plain wall behind her. |
| She runs, then stops, then turns and cries while the camera zooms out | Four actions and two movements in one shot | Split into three shots of two or three seconds, one action and one move each. |
Say what is there, not what is not.
Make it in SceneMixer, start to finish
- Script: paste your own on Upload Script, or write the brief above in AI Create and let it draft the episodes.
- Step 1: Type Short Drama; Scene Style Neon Nights, American Small Town; Costume Style Soft Everyday, Streetwear; Visual Style Digital HD, Shot on Film; Target Region; dialogue language; 9:16 Portrait. Then read the script.
- Step 2: Paste the three character prompts into the appearance of each character card and the two place prompts into the scenes; the shot prompts go into the storyboard. The first 15 setup images on an account are free.
- Storyboard: generate episode 1 first (free on your first drama) and read it before you render anything. One place per segment.
- Video: render on the default tier at 10 credits a second, fix single shots instead of whole episodes, then put the episode together and post.
| Prompt for | Write | Avoid |
|---|---|---|
| A shot | Framing, one camera move, who and where, one action, sound, seconds | Tags, stacked actions |
| A character | Face, anchor, build, one outfit | Beautiful, charismatic, neutral expression |
| A place | Where the door, window and light are | No furniture, no people |
| An ad | A photo plus what it is and who it is for | Best, number one, guaranteed |
| A line | Name (tone): the line, then who voices it | More words than the shot has seconds |
FAQ
What is a good AI video prompt?
One shot: framing, one camera move, who and where, one action, sound and seconds, in plain sentences.
Do I need magic words like cinematic or 8K?
No. They give the model nothing to place, push useful words down, and sometimes get painted literally. Pick the look with the style preset.
How long should a prompt be?
A shot is one or two sentences; a character description is a short paragraph; an ad is one line plus a photo.
Should I list what I do not want?
No. Naming something invites it. Say what is there instead: an empty street, not no crowd.
Can I reuse these prompts in any tool?
They are model-neutral and written as ordinary sentences, so they carry over; only the @Name tags and the voice marker are SceneMixer's.
Where do I paste a prompt in SceneMixer?
Shot prompts go into the storyboard text; character and place prompts go into the descriptions in Step 2; an ad takes a photo and one line.
How do I write a character image prompt?
Face first, with one unmistakable anchor, then build, then one outfit. Leave the lighting and finish to the sheet.
Does a longer prompt make a better video?
Not by itself. Extra detail helps only if the model can draw it; extra actions hurt.
Is it free to try these?
Writing and editing the storyboard is free; new accounts start with 50 credits and the first 15 setup images are free.
Updated November 2, 2026