DRAFT PREVIEW: not launched.
DRAFT FOR REVIEW · Broadside Code is a placeholder

Free · Shot Library

Music Video Shot Library: 10 Seedance prompt patterns

Ten shot types every music video needs, each with the model to use, a copy-paste pattern and the rules that keep it from failing. From the method I use on 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.

The shots

  1. 1. Establishing wide
  2. 2. Instrument performance close-up
  3. 3. Locked singing shot
  4. 4. Dialogue or interview close-up
  5. 5. Crowd and stage wide
  6. 6. B-roll insert
  7. 7. Walking shot
  8. 8. Head turn or look-away
  9. 9. Drive shot
  10. 10. Candid behind-the-scenes still
One rule under all ten: pixels beat words. 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, set the place, carry mood with no close face.
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 is theater. 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 2.5Lip sync that holds.
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. Dialogue or interview close-up

Seedance 2.5One character, one or two short lines.
REGISTER: <one word>.
VOICE: <approved voice text, verbatim every time>. <Nationality.> Every sentence ends low, pitch falls at the end of each line and never rises.
TASK: <the beat as prose; name the brow; an eyeline into the other person's eyes; one piece of business>.
"<line with tics inside the quotes, ellipses for pauses>"
CAMERA: <framing, one motivated move or locked>.
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. The line spoken a single time, a beat of silent room tone before and after. No echo, no reverb, no repeated words, no doubled audio, no second voice, no trailing repeat.
NEGATIVES: no music, <only things that could appear>
  • The mouth stays shut for about the first two seconds. First important word at 3 s or later.
  • Duration: words ÷ 2.5, plus 1.5 s per pause, plus 1 s each end.
  • Two women talking: one roll each.

5. Crowd and stage wide

Seedance 2.0The live-show feel, 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.
  • Wherever the lead walks, there is life. Place it 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>. Image 2 is <character>, the only face authority. Image 3 is a three-quarter reference.
<Task: one beat: <character> turns from <A> to look at <B>.>
CAMERA: one simple move, the character steady on screen.
  • Seedance blocked four photoreal head turns away from camera. H3 rendered one and held the face.
  • 5 s minimum: one beat per roll, keep the best 2 to 3 seconds.
  • H3 is near silent. 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.>
  • The staging is the tell. 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.

[PLACEHOLDER: "Updated weekly" line and last-updated date, once the refresh routine is running.]