Music video shot library: 10 Seedance prompt patterns
Ten common music video shots, each with the model to use, a copy and paste pattern, and the rules that keep it from failing. Free to read and copy. From the method behind the Alder videos.
These are patterns, not finished prompts. Fill the angle brackets with your own artist, room and sound. The rules under each one come from real rolls on the Alder videos. Your filled in prompt is yours to test: roll at 480p first.
One rule under all ten: reference images beat prompt text. The model copies your reference images literally. Lock faces, clothes and props with images, keep the prompt for acting and sound, and keep it under 200 words.
1. Establishing wide
Seedance 2.0Open a section and show the place, with no face close to camera.
Image 1 is the <location plate> for the room, the light and the camera angle only.
ERA: <one line>.
LOCKS: <only what breaks the shot if wrong>.
<Task, three or four sentences: what is happening in the place, in story order, ending in one end state. Every animal and prop gets its own small motion.>
CAMERA: a slow <push / drift / pan>, <bounded: a quarter turn at most>.
Sound: only what a microphone at <the place> would record. <real sounds>. No music anywhere in the clip, from the first frame to the last.
NEGATIVES: no music, <the stock default you keep getting>, <up to eight more>
Seedance 2.0 is the default for anything where no face is close.
Plates for every shot, wides included. A wide with no plate gets the stock default.
Keep people over about 100 px tall at 480p, or the blockout shows through.
2. Instrument performance close-up
Seedance 2.5 + your own guideHands that follow the real recording.
Image 1 is the <plate> (start image). Image 2 is <player>, the only face authority. Image 3 is <the instrument> for the instrument only.
REFERENCE ROLES: The <length> mannequin video is the authority for <arm, wrist, finger and bow / pick> movement only. Replace the mannequin with <player>.
TIMING: Render only reference-video seconds <start> through <end>, at the original 1x speed, as output seconds 0.000 through <length>. Do not condense, skip ahead, restart the movement, add a lead-in, or retime the strokes. Begin already playing on the first frame.
<Eyes on the <fretting hand / bow>, a focused, alive face, never neutral.>
Sound: only what a microphone at <the place> would record. <room tone, the pick on the strings>. No music anywhere in the clip, from the first frame to the last.
NEGATIVES: no music, <only things that could appear>
A prompted instrument roll only mimes playing. The guide does the playing; build it from the recording.
Never a lone raised finger on the plate or the guide. It reads as a middle finger and trips the filter.
Strip the model's audio and lay your own recording in post.
3. Locked singing shot
Seedance 2.5Singing to camera, with lip sync.
Image 1 is the <singing plate>, mouth open mid phrase. Image 2 is <singer>, the only face authority.
Audio: the dry vocal WAV for this phrase (15 s max) as the audio reference.
<Task: <singer> sings the phrase, dynamics following the phrase, a breath between lines.>
CAMERA: locked, flat on, <medium close-up>.
Sound: only what a microphone at <the place> would record. <room tone>. No music anywhere in the clip, from the first frame to the last.
NEGATIVES: no music, <only things that could appear>
Attach the vocal, then run lip sync with the same WAV.
Singing shots stay locked. No profile singing; flat on or three-quarter at most.
Never small in a wide for lip sync. Open-mouth singing needs a singing plate.
4. Talking clip
Seedance 2.5One character, one to three short lines.
An AUDIO REFERENCE is attached: a clean clip of <character> speaking, the exact and only authority for <CHARACTER>'S VOICE. <CHARACTER>'S VOICE: <age>, <accent named precisely>, NOT <the accents it drifts to>. Plain natural speech, not a performance. Only one person speaks in this roll.
<Place>. <Character> alone, <framing>, exactly as the start_image, camera locked off, one continuous shot. <Locks with exact counts; colours with what they are not.> She looks <screen side> into the other person's eyes.
Task, <length> seconds. 0 to 0.5 s: she holds the look. About 0.5 s: <one small action>: "<line, with the uh inside the quotes>" <a beat, a different action>. About <n> s: "<next line>" Her mouth is still for the last two seconds.
SOUND: documentary sound only, the room tone of <the place>, nothing else. NO MUSIC of any kind, no score, no band.
Negatives: no music, no second speaker, nobody else talks, no readable signs, no logos, no camera movement
Open on the held start pose, or the model may skip your start frame.
A voice sample from the character's own approved audio held the accent where adjectives failed. Never use one with two speakers.
Two silent seconds at the end stop the last word clipping.
5. Crowd and stage wide
Seedance 2.0A stage and crowd, with crowd sound that sounds like people.
Image 1 is the <venue plate> for the room, the light and the camera angle only.
<Task: the band plays up front; the crowd <beat-by-beat moments>.>
CAMERA: <one bounded move>.
Sound: only what a microphone at <the venue> would record. Hundreds of individual people yelling, each voice distinct and sharp. No reverb, no echo, no swelling. No music anywhere in the clip, from the first frame to the last.
NEGATIVES: no music, <the wrong default, e.g. no seated audience>
Never write wash, swell, wave or hold for crowd sound. A "wash" came back underwater.
No reference means the stock default. Name the wrong default in the negatives.
Lay the song in post.
6. B-roll insert
Seedance 2.0Props, hands, details between performance shots.
Image 1 is the <plate> for the frame. Image 2 is <the prop> for the prop only.
<Task: one small action on the prop, one end state.>
CAMERA: <a slow push or a reason it is locked>.
Sound: only what a microphone at <the place> would record. <the prop's own sounds>. No music anywhere in the clip, from the first frame to the last.
NEGATIVES: no music, <only things that could appear>
Inserts run 1.5 to 2.5 seconds in the cut. Roll with half a second of handle each end.
Zoom the hero prop in the plate first. A glass rim read as broken in three rolls.
Count props like fingers; each one exists exactly once.
7. Walking shot
Seedance 2.0A character moving through a space.
Start frame: <character> at <point A>. End frame: <character> visibly further along at <point B>.
<Task: <character> walks to <B>; bystanders each have a starting activity, a reaction, and a settle.>
CAMERA: <follow or hold; no tracking at walking speed>.
Sound: only what a microphone at <the place> would record. <footsteps, the room>. No music anywhere in the clip, from the first frame to the last.
NEGATIVES: no music, feet planted, <up to eight more>
If the start and end frames are at the same spot, walkers tread in place.
Put background people and movement wherever the lead walks. Place them in the animatic.
"Feet planted" goes in the negatives.
8. Head turn or look-away
MiniMax H3The shot Seedance keeps refusing.
Image 1 is the <plate>, for face, hair, clothes, framing and room. Image 2 is <character>'s head card, the only face authority.
<Look locks.> The <target> is behind her, on the <RIGHT> side of the frame. She turns her head to camera <right>, over her <left> shoulder, only partway, her body staying where it is. She never turns toward camera <left>; camera <left> is where <the other person> sits. Then her head comes slowly back around.
Negatives: no turning toward camera <left>, no turning the whole body, no zoom, no camera movement
Seedance blocked four photoreal head turns away from camera. H3 rendered the same prompt and held the face.
"Over her shoulder" has two answers. Name the screen side and where the other person sits.
5 s minimum and near silent: keep the best 2 to 3 seconds and lay room tone under it.
9. Drive shot
Seedance 2.0Drives, cars, night-road B-roll.
Image 1 is the <car interior or exterior plate>. Image 2 is <driver>, the only face authority.
<Task: <driver> drives, <one piece of business>; the road and lights move past.> The driver sits on the <left> side.
CAMERA: <locked to the car, or one bounded move>.
Sound: only what a microphone in <the car> would record. <engine, tyres, the road>. No music anywhere in the clip, from the first frame to the last.
NEGATIVES: no music, <only things that could appear>
Name the driver's side and let the model do the mechanics.
Seedance 2.0 is the default for drives.
Pairs with a neon night-drive loop for Canvas and Reels.
10. Candid behind-the-scenes still
Image model (Nano Banana 2.1)A still that looks like someone wandered in and took it.
A stranger's phone snap of <the band> in <the place>: off-axis, chest height, cropped limbs, uneven spacing. <Character references, one job each.> <Light that is really in the room.>
Posed staging gives it away. Grain and tilt do not fix a posed frame.
Stills are cheap: fix freely before any video roll.
Fix proportion by editing the pixels, not with words on a regeneration.
Get the PDF version
10 Seedance Music Video Prompt Patterns: each pattern with the failure it fixes and the Alder example behind it. Free.
Want the whole method? The Seedance Music Video Playbook covers models, characters, dialogue and guide driven instrument performance.