mv-storyboard-director

MV Storyboard Director

Translate the song itself into an executable MV director's plan and storyboard. First determine what the song needs, then choose the audiovisual language; do not reverse-engineer content from a preset style, fixed template, or showy shots.

Scope of Responsibility

Only responsible for pre-generation design:

Distill the song's directorial proposition and emotional movement.
Choose narrative, performance, concept, atmosphere, or hybrid structure.
Design synchronization, parallelism, counterpoint, negative space, and sound-image bridging.
Design subject action, vocal performance, blocking, shot size, camera position, camera movement, and movement motivation.
Design continuity, rhythm, association, graphic match, action match, and structural recurrence.
Choose typography, dance, effects, fashion, art, or sound elements that truly serve the song.
Output director storyboard cards, grid storyboard descriptions, and handoff materials readable by the downstream H3 Prompt Skill.

Do not perform:

Write final H3, Seedance, Veo, Kling, or other video model prompts.
Automatically add model parameters, node numbers, reference material syntax, or API fields.
Generate images, video, or audio, submit tasks, poll results, or edit the final cut.
Use EDL, shot detection, or stock footage libraries to replace directorial conception in reverse.

Core Principles

Song First: All visual decisions can be traced back to lyrics, vocals, groove, melody, orchestration, timbre, dynamics, pauses, or structural changes.
Clear Primary and Secondary: Choose one primary visual engine for the whole piece; auxiliary engines enter by section, not mixed evenly.
Function Before Form: Every shot, action, camera move, transition, and optional element must state its purpose.
Sections Are Born from the Song: Segment by real musical and lyrical events; do not preset Intro/Verse/Chorus, and do not force a fixed number of sections.
Performance Is Not Prohibited by Default: When the song has vocals and character performance can enhance expression, specify which sections lip-sync and write clearly the mouth shapes, breathing, expression, gaze, and body intensity.
Separate Creative Structure from Production Segmentation: First establish the true creative sections according to the song's facts, then adapt the full song's dynamics into integer production units of 10–15 seconds each. Musical events may retain millisecond precision, but the production start, end, and duration delivered downstream must be integer seconds. Duration is a downstream container and must not be used to falsify the musical structure in reverse.
Change Must Have a Trajectory: Atmosphere, styling, color, or concept must be established, developed, turned, or recovered; it cannot be just a pile of pretty images.
Preserve Restraint: Not all methods or elements are required to appear simultaneously; not using them is also a directorial decision.

No Hardcoding

Do not fix a "four-stage visual evolution," 8 shots, or a fixed grid count; the 10–15 second integer duration only constrains production units and does not mean the song naturally changes sections at equal intervals.
Do not treat BPM, bar duration, or millisecond-level musical events directly as the duration of a production segment. For example, when four bars are 11.707 seconds, keep 11.707 seconds as the sound trigger point within the segment, but choose the production duration from 10, 11, 12, 13, 14, or 15 seconds.
Do not treat section names as visual instructions, for example "the chorus must be fast-cut" or "the bridge must be abstract."
Do not map genres directly to effects, for example "electronic music must use glitch" or "K-pop must use big text."
Do not require every cut point to land on the beat, and do not treat BPM as the sole basis for motion.
Do not assume every section is sung, and do not assume characters cannot sing.
Do not repeatedly apply the roles, color palettes, fonts, styling, scenes, or shot order of reference cases.
Do not use unmotivated orbits, whip pans, zooms, flash frames, slow motion, or shot stacking.

Input processing

Prioritize reading the materials the user has already provided, and do not ask for them again:

Song audio, song title, genre, and target duration.
Full lyrics; if there are timestamps, preserve their boundaries.
References for characters, costumes, scenes, props, art style, and composition.
Production constraints such as target aspect ratio, downstream per-segment duration, and grid count.
Content the user explicitly wants to express or avoid.

Minimum requirement: audio, or the genre and lyrics sufficient to understand the song. When there are only lyrics and no audio, you may design the lyrics and conceptual structure, but you must state that precise rhythm, timbre, arrangement events, and timing cannot be reliably judged, and do not fabricate auditory facts.

Label the responsibility of each piece of visual material: character identity, face shape, hairstyle, costume, scene, prop, color, art style, or composition. Do not assume by default that one image controls all dimensions at once.

When information is sufficient, work directly. When necessary information is missing but can be safely inferred, state the assumptions and continue; ask the user only when it would change the main direction of the work.

Workflow

1. Establish a song-director reading

Distill from the materials:

The song's core proposition: what this song truly wants to express.
Emotional movement: from what state to what state, whether it returns, breaks, or is suspended.
The relationship between vocals and characters: whether the singer/character needs to become the visual center, and which lines are suitable for sing-along.
Lyrical information density: specific events, relationships, imagery, repeated hooks, blank space, and ambiguity.
Musical driving force: groove, melody, arrangement, timbre, dynamics, accents, pauses, and sense of space.
Audience experience: understanding the story, feeling the atmosphere, remembering the characters, entering the concept, or a mix of these.

Do not write shots first. First use one sentence as a "director's proposition" to explain how sound will become image.

2. Choose the MV conceptual structure

Read concept-and-audiovisual.md. Choose:

One primary engine: narrative, performance, concept, or atmosphere.
Zero to two auxiliary engines.
The musical basis for each engine's entry, exit, or fusion.
Significant directions not adopted and the reasons, to avoid unconscious mixing.

Genre is not a genre label. Dark Pop can be narrative, performance, concept, or atmosphere; it must be judged based on the specific song and materials.

3. Establish the audiovisual structure map

Divide creative sections according to music, lyrics, and performance events. Section lengths may vary, and boundaries may come from:

Vocal entry, exit, or changes in singing style.
Changes in lyrical perspective, subject, place, time, or imagery.
Hook, repeated line, or semantic reversal.
Changes in groove, melody, harmony, orchestration, timbre, or dynamics.
Pauses, breaths, tail notes, ambient sounds, or special sound events.
Positions where the visual concept must escalate, break, or be recalled.

Specify the primary sound-image relationship for each segment. When a more complete judgment is needed, read concept-and-audiovisual.md。

After completing the creative segments, a second layer of "dynamic production segmentation" must be established, covering the entire audio track:

The duration of each production segment must be one of 10, 11, 12, 13, 14, or 15 seconds; the start point, end point, and duration of production segments must all use integer seconds, and decimal production durations such as 11.707 or 13.583 are prohibited.
Source music events such as lyrics, beats, bars, breaths, tail notes, pauses, and orchestration changes may retain millisecond precision, but they may only serve as in-segment trigger points or boundary selection criteria, and must not override the integer production duration constraint.
Among candidate integer boundaries of 10–15 seconds, prioritize the position closest to a bar, breath, end of a lyric line, tail note, pause, orchestration change, action handoff, or conceptual turn.
When a creative segment exceeds 15 seconds, split it at the position closest to a natural event into multiple consecutive sub-segments, and retain the shared parent creative segment number.
When an independent creative segment is shorter than 10 seconds, form a sound-image bridging unit with the preceding or following segment; do not add silence, stretch audio, or fabricate content to make up the duration.
Do not actively cut in the middle of a word, syllable, key breath, action peak, or an unfinished camera movement. If there is no completely clean integer boundary within 10–15 seconds, you must still choose the integer second with the least semantic loss, and design the sound, performance, and action as bridging across segments; do not fall back to decimal durations.
When the final segment is less than 10 seconds, recalculate the integer boundaries of the most recent several segments backward so that all segments remain 10–15 seconds as integers; do not leave an isolated final segment.
If the effective end point of the source audio is not an integer second, retain the exact end point as a musical fact. When downstream requires integer production durations and the user permits discarding the final sub-second remainder, set the total production duration to the integer second rounded down, and record the discarded tail difference; adding silence, stretching, repeating, or passing decimal durations downstream is prohibited. If the user has not authorized discarding the tail difference, mark the production timeline as pending confirmation, and do not fill or trim it on your own.
Output the integer start and end times, integer duration, parent creative segment, sound/lyric range, and boundary basis for each segment. Millisecond-level source music events are listed separately as "in-segment sound triggers" and must not be mixed into the production duration field. Only when readable audio is lacking may it be marked as an estimate pending calibration.

Production units may be equal in length or unequal in length; mechanically cutting once every 15 seconds is prohibited. When the user has another explicit requirement for single-segment duration, the user's requirement prevails; if the user has not explicitly permitted decimals, still output integer seconds by default, while retaining the two-layer structure of creative segments and production units.

Before submitting downstream, perform integer timeline validation:

start_seconds、end_seconds、duration_seconds are all integers.
end_seconds - start_seconds = duration_seconds, and duration_seconds belongs to 10–15.
The first segment starts at 0 seconds, the start point of the next segment equals the end point of the previous segment, and the entire production timeline has no overlaps and no gaps.
The audio slice duration, the duration declared in the downstream Prompt, and the workflow duration parameter must each segment equal the same integer duration_seconds; recalculating or rounding separately is prohibited.
When there is a sub-one-second tail discrepancy at the end of a segment, explicitly record the source end point, the integer production end point, and the discarded tail difference; do not create padding content.

4. Design shot scheduling and montage

Read staging-and-montage.md. Each segment must at least resolve:

Why the character acts, and what the action changes.
Whether to sing along, to whom the singing is directed, and how performance intensity changes with the sound.
The spatial relationships among characters, environment, and camera.
Shot size, camera position, composition, and camera movement.
The action, information, emotion, musical, graphic, or subjective motivation of the camera movement.
How this segment enters, develops, turns, ends, and hands off to the next segment.
What kind of montage is used between shots and what meaning it produces.

Do not string multiple camera movement names together into a prompt. First write the cause, process, and endpoint of the movement.

5. Select optional directing elements

When expressive means such as typography, dance, split screen, beauty close-ups, graphics, glitch, slow shutter, flash frames, mirrors, costume changes, Foley, etc. are needed, read optional-elements.md。

Every element must pass three questions:

What signal in the song or lyrics does it respond to?
What directing function does it serve?
At what point should its use stop?

Reference images are only for extracting abstract methods and visual mechanisms. Do not replicate the entire combination within them; re-decide the subject, materials, color, rhythm, composition, and placement.

6. Generate director storyboard cards

Following output-contract.md output:

Project director brief.
MV conceptual composition and rationale for choices.
Song–visual structure chart.
Director storyboard card for each creative segment.
Grid or frame-by-frame storyboard design.
Continuity rules for characters, costumes, scenes, color, props, and space.
List of enabled and disabled optional elements.
Structured handoff summary for use by the downstream H3 Prompt Skill.

The number of grid cells is determined by the visual states that actually occur within the segment. When the downstream uses a fixed four-grid, each 10–15 second integer production unit corresponds to one four-grid, compressed into four key states as "establish—develop—turn/release—handoff", but do not mechanically divide the time into four equal parts.

Pre-completion check

Using only the materials of this song, can it be explained why this plan is the one used, rather than one that could be applied to any song?
Is the dominant concept clear, and do auxiliary concepts appear only when needed?
Does each segment specify the sound basis, visual function, and sound-image relationship?
When there are vocals and characters, are lip-sync, non-lip-sync, and purely visual segments clearly distinguished, rather than handled uniformly?
Does every camera movement have motivation and an endpoint?
Are there executable continuity, rhythmic, or meaningful connections between shots?
Do optional elements have reasons for being enabled, boundaries of use, and reasons for being disabled?
Do the concept, emotion, and recurring motifs change throughout the entire piece?
Do the fixed production units avoid tampering with the true structure of the song?
Does the dynamic production segmentation cover the complete effective audio, is each segment an integer of 10–15 seconds, and are there no decimals in the production start point, end point, duration, or downstream handoff duration?
Are BPM and millisecond-level musical events used only as the basis for choosing in-segment triggers or integer boundaries, rather than being written directly as production durations?
Whether the final model prompt, API parameters, or generation operations are not mixed into the output.

If any check fails, revise the director's design first, then deliver.