Prompt Engineering Guide

Write Prompts That Direct Your Music Video, Shot by Shot

An AI music video is only as good as the prompt behind it. This is the practical guide to describing a song like a director — the shot formula, the camera language, how to sync cuts to the beat, and how to keep your singer looking like the same person from the first frame to the last.

8
prompt building blocks
20+
camera & shot terms
3
full worked examples
Mindset

Stop Writing Captions. Start Writing Shot Lists.

The single biggest upgrade to your AI music videos isn't a better model — it's a better prompt.

Most people describe a song the way they'd write a caption: "a sad girl in the rain, cinematic." The model then has to guess everything that matters — who she is, what she does, where the camera sits, how the light falls, when the cut lands. A music video prompt works the opposite way. You hand the AI a shot list: a specific subject, a precise action, a framed camera move, a defined look, and a constraint or two. You stop hoping and start directing.

Caption — the model guesses

A sad girl in the rain, cinematic, emotional, beautiful, 4k

Shot list — you direct

Medium shot, a young woman in a soaked grey coat stands under a streetlight, slowly lowering her head as rain runs down her face. The camera pushes in gently to a close-up. Cool blue night light, soft reflections on wet asphalt, cinematic film look. No on-screen text.

Same idea. One leaves the result to chance; the other gets the shot you pictured.

The Formula

The Anatomy of a Shot Prompt

Eight building blocks, in roughly this order. You don't need all eight every time — but the more you lock down, the less the AI improvises.

Subject + Action + Scene + Light & Color + Camera + Style + Quality + Constraints

1 · Subject

Who or what is on screen. Pin it down with two or three stable visual traits — clothing, hair, age, type — so the AI can recognize it and hold it steady across shots.

a young woman in a red silk dress and a straw hat

2 · Action

What the subject does, described by body part and degree. "Raises her hand" beats "moves." Favor slow, continuous, connected motion over wild action — it renders far cleaner.

slowly lifts her chin and turns her head toward the window

3 · Scene

Where it happens and what surrounds the subject — location, time of day, the foreground and background elements that build the world.

a neon-lit rooftop corridor at night, hologram ads flickering behind

4 · Light & Color

The mood lives here. Name the light direction, the warmth, and the palette — the same scene reads as romance or dread depending on this one line.

warm golden-hour backlight, soft shadows, amber and teal palette

5 · Camera

How it's shot: the shot size and one camera move. This is the difference between a flat clip and a directed frame. Keep it to a single move per shot.

slow push-in from a wide shot to a close-up

6 · Style

The overall art direction — live-action film, 2D anime, 3D animation, ink-wash, claymation, pixel art. State it explicitly, or a realistic reference photo will quietly drag everything toward live action.

2D anime style, hand-drawn, cel shading

7 · Quality

The finish: clarity, texture, and the cinematic qualities you want baked in. A short, standard phrase is enough.

high definition, rich detail, filmic texture, natural color

8 · Constraints

What you don't want. The most useful negatives keep stray text, watermarks and logos out of frame, and stop the look from drifting.

no subtitles, no watermark, no logo

Camera Language

Speak the Camera's Language

AI video models respond precisely to standard film terms. Learn this vocabulary and you can frame any shot in a few words. The golden rule: one camera move per shot — stacking push, pan and zoom into one line makes the picture unstable.

Camera moves

Push in

Camera moves toward the subject — builds intimacy or tension. The workhorse move for a chorus close-up.

Pull back

Camera retreats to reveal context — great for a reveal, or to let a scene breathe at the end of a line.

Pan

Camera pivots left or right from a fixed point — sweep across a stage, a crowd, or a skyline.

Tilt

Camera pivots up or down — rise from feet to face, or drop from sky to street.

Tracking / dolly

Camera travels alongside the subject — walk-and-talk energy that keeps a moving singer in frame.

Orbit

Camera circles the subject — a 180° or 360° arc that shows them off. Perfect for a hook.

Crane / boom

Camera rises or descends through space — lift to a wide for scale, or descend into the action.

Zoom

Focal length changes to magnify or compress — snappy zooms punctuate a beat; slow zooms add unease.

Shot sizes

Extreme wide

Subject small in a big world — sets place and scale, the establishing shot of a song.

Wide / full

Whole body in frame — the best size for choreography and full-body motion.

Medium

Waist up — the conversational default; reads body language and face together.

Close-up

Face fills the frame — emotion, lip-sync, the money shot of the chorus.

Detail / macro

A single detail — eyes, hands on strings, a falling tear — for rhythm and texture cuts.

Angles & perspectives

Eye level

Neutral and grounded — the audience meets the subject as an equal.

Low angle

Camera looks up — makes the subject powerful, heroic, larger than life.

High angle

Camera looks down — makes the subject smaller, vulnerable, or lost.

Over the shoulder

Frames past one figure toward another — connection, dialogue, tension between two.

POV / first person

The subject's own view — pulls the viewer inside the moment.

Aerial / drone

High above, looking down — vast landscapes, crowds, the epic outro.

Pro tip: one move per shot. Want a push and an orbit? Make them two shots and cut between them — the cut is part of the music.

Motion & Emotion

Make It Move — and Make It Feel

How you describe action decides whether a shot looks alive or like a flickering photo. Four habits separate clean motion from a melting mess.

Be specific by body part

Name the part and the degree: "slowly raises her right hand," "tilts her head a few degrees," "plants her foot hard." Vague verbs like "moves" or "dances" invite chaos.

Favor slow, continuous motion

Gentle, connected micro-movements render beautifully; explosive sprints, big jumps and violent spins tend to warp. Save the high energy for cuts, not single shots.

Write the transition

Connect one action to the next so motion has momentum: "using the turn's momentum, she lifts her arm," rather than two unrelated poses.

Externalize the emotion

Don't write "she's sad." Write the body that shows it. The model can't render an adjective — it can render a trembling shoulder.

Turn feelings into footage
Sadness
head lowered, shoulders trembling slightly, eyes glistening, fingers clutching a sleeve, a tear welling but not falling
Joy
an uncontainable smile, brows lifting, a lighter step, spinning once on the spot
Tension
glancing at the door, fingers drumming, shallow breathing, eyes darting
Anger
fists clenched, jaw set, chest heaving, a gaze sharp as a blade
Relief
a long exhale, shoulders dropping loose, a faint smile, eyes lifting to the distance
Storyboard & Sync

Turn a Song into a Shot List

A music video isn't one prompt — it's a sequence. Break the song into sections and give each its own shot. Think on a timeline: who, where, doing what, and how the camera moves, in the order it happens.

The most reliable way to prompt a full video is to label your shots — Shot 1, Shot 2, Shot 3 — and describe them in order. Map them to the song's structure so the picture turns when the music turns.

01

Intro

Set the world. An establishing wide or a slow detail — let the viewer arrive before the vocal does.

02

Verse

Tell the story. Medium shots, walk-and-talk tracking, the subject moving through the scene.

03

Build / pre-chorus

Raise the tension. Push in, tighten the framing, quicken the cuts as the energy gathers.

04

Chorus / drop

The payoff. The close-up, the orbit, the boldest light and the biggest move — the shot people remember.

05

Bridge / outro

Release or resolve. Pull back to a wide, fade the light, let the last image breathe.

Every shot, four layers

Camera

The move or the cut — "slow push from a wide," "fixed frame," "cut to a close-up."

Subject & expression

The key action and the change in expression for this beat.

Space

Where the subject is and how the scene around them shifts.

Sound & beat

What lands on this shot — the downbeat, the drop, a lyric hitting home.

Sync cuts to the beat, not the clock. Describe what happens in order and let the cuts land on the rhythm — pinning exact seconds ("0–3s") tends to confuse the model and break the result.

Character Consistency

Keep Your Singer the Same Person

Nothing breaks a music video like a face that changes every shot. Reference images and disciplined prompting keep your character locked from first frame to last.

Lock identity with 2–3 traits

Describe the subject with a few stable, unmistakable features — outfit, hair, a defining accessory. Reuse the exact same description every time you mention them.

Use a close-up plus a full body

A clean head-and-shoulders portrait pins the face; a full-body shot pins the outfit and proportions. Together they hold identity far better than a single image.

Put the important reference first

The earlier a reference appears, the more weight it carries. Lead with the one that matters most for this shot.

Don't overload the references

More images isn't better. Too many competing references blur the AI's sense of what matters — a couple of strong, clear ones beat a pile of weak ones.

In SunoMV, save your character and locations as scene references, then point each shot's prompt at them — so the same singer carries through every line of the song without re-describing them each time.

Avoid These

The Mistakes That Ruin AI Music Videos

Almost every disappointing result traces back to one of these. Spot them early and your hit rate climbs fast.

Identity drift

The face changes between shots. Fix it with a clean close-up reference, a consistent 2–3 trait description, and the key reference placed first.

Unwanted text on screen

Random captions or gibberish appear. Add "no subtitles, no text," and prefer widescreen — vertical frames produce stray text more often.

Style drift

An anime idea renders as live action. State the style explicitly and, for a strong look, start from a reference image already in that style.

Accidental twins

The same person appears twice in frame. Tie each subject to its own reference and add a constraint against duplicate, identical-looking people.

Too much motion

Fast, violent action warps faces and limbs. Break big movement across cuts and keep each single shot calm and continuous.

Reference overload

Five competing images muddy the result. Use a couple of strong references and let the prompt do the rest.

Pinning exact seconds

Hard time codes ("at 2s, cut") destabilize generation. Describe order and let the cuts ride the beat instead.

Mixing languages mid-line

Switching languages inside one spoken line confuses pronunciation. Keep each line in one language; common words render most cleanly.

Worked Examples

Three Prompts, Start to Finish

Copy the shape, swap in your song. Each example is a multi-shot prompt you could paste today — built from the blocks above.

Melancholic indie ballad

Intimate, rain-soaked, one character, slow and emotional.
Style: cinematic live-action, cool film look. Subject: a young woman in a soaked grey coat, dark hair stuck to her face. Shot 1 — wide establishing shot, an empty street at night under amber streetlights, rain falling, she stands small in the frame; slow push-in. Shot 2 — medium shot, she slowly lowers her head as rain runs down her face, breath visible. Shot 3 — close-up, a single tear wells but doesn't fall, her eyes lifting toward the light. Cool blue palette, soft reflections on wet asphalt, high definition, filmic texture. No subtitles, no watermark.

Why it works: one locked character, three escalating shot sizes, a single move per shot, emotion shown through the body, and a clean constraint line.

Upbeat synth-pop

Bright, kinetic, color-popping, built around the chorus drop.
Style: vibrant 3D animation, glossy and colorful. Subject: a dancer in a neon tracksuit and white sneakers. Shot 1 (verse) — medium tracking shot following her down a sunlit corridor of color-blocked walls. Shot 2 (build) — low angle, push in as she crouches, the light pulsing brighter. Shot 3 (chorus drop) — 360° orbit around her as she throws both arms up, confetti bursting, saturated pink and cyan light. Snappy, energetic, high definition. No on-screen text, no logo.

Why it works: shots are mapped to song sections, the biggest move and brightest light are saved for the drop, and the energy lives in the cuts rather than one frantic shot.

Cinematic cyberpunk / hip-hop

Stylized, neon, moody, larger than life.
Style: 3D cyberpunk sci-fi animation, cool blue-violet palette. Subject: an artist in a black hooded jacket with glowing trim. Shot 1 — aerial drone shot descending toward a rain-slicked neon city at night. Shot 2 — low-angle medium shot, he walks toward camera through hologram ads, head rising to meet the lens; slow tracking. Shot 3 — close-up, reflections of neon sliding across his face as he mouths the hook; fixed frame. Moody volumetric light, deep shadows, filmic grain. No subtitles, no watermark, no logo.

Why it works: a strong stated style holds the whole sequence, the low angle builds presence, and each shot uses exactly one move — descend, track, hold.

Cheat Sheet

The One-Screen Summary

Everything above, condensed. Screenshot it and keep it next to your prompt box.

Subject + Action + Scene + Light & Color + Camera + Style + Quality + Constraints

Do

  • Describe a shot list, not a caption.
  • Lock your subject with 2–3 stable traits and reuse them.
  • One camera move per shot — cut for the rest.
  • Show emotion through the body, not adjectives.
  • Map shots to the song's sections; sync cuts to the beat.
  • End with constraints: no text, no watermark, no logo.

Don't

  • Don't stack push + pan + zoom into one shot.
  • Don't pin exact time codes.
  • Don't ask for violent, explosive motion in a single shot.
  • Don't overload the prompt with competing reference images.
  • Don't leave the style unstated next to a realistic reference.
  • Don't switch languages inside one spoken line.

Music Video Prompt FAQ

What is music video prompt engineering?

It's the craft of describing a song to an AI video model so it returns the shots you actually pictured — choosing the subject, action, camera move, lighting and style for each part of the track, instead of typing a vague description and hoping. Done well, it's the difference between a generic clip and a directed music video.

Do I need to know film terms to write good prompts?

It helps, and they're quick to learn — push in, pan, close-up, low angle. The vocabulary on this page covers most of what you'll ever need. AI models respond to these standard terms precisely, so a few words frame a whole shot.

How do I keep the same character across the whole video?

Use a reference image — ideally a clean close-up plus a full-body shot — and describe the character with the same 2–3 traits every time. In SunoMV you can save a character as a scene reference and point every shot at it, so they stay consistent from the first line to the last.

Why does my AI music video have random text or subtitles?

Models sometimes invent on-screen text. Add a constraint like "no subtitles, no text" to your prompt, and prefer a widescreen frame — vertical videos tend to produce stray text more often. You can also crop to vertical afterward.

Should I write one long prompt or several shots?

Several shots. Label them Shot 1, Shot 2, Shot 3 and describe them in order, mapped to the song's sections. A sequence of clear shots is far more reliable than one paragraph trying to do everything at once.

Why does the action look melted or warped?

Usually too much motion in a single shot. Favor slow, continuous, body-specific movement and break big action across cuts. Calm shots cut together read as high energy; one frantic shot just warps.

Can I use these techniques with any AI video model?

Yes. The formula — subject, action, scene, light, camera, style, constraints — is universal craft. SunoMV lets you pick from several advanced video models and switch freely, but the way you write the prompt carries across all of them.

Where do I actually do this in SunoMV?

Paste a song or create one, then open the song page: each lyric line becomes a shot you can prompt individually, with scene references to keep characters and locations consistent. Title, cover, and author live under Edit Info (pencil). Start free and direct your first video below.

Now Go Direct Your Song

You've got the formula, the camera language and the examples. Paste a track, write your first shot list, and watch the AI shoot it.