Communitygithub.com

audio-driven-pmv-typography-skill

Plans, designs, edits, and renders polished PMV typography, kinetic typography, lyric-style, and audio-driven motion-graphics videos. Use whenever creating a video where text, photos, illustrations, footage, or graphic elements are edited and animated to music, voice, or other sound. Supports text-only, photo-led, mixed-media, and abstract visual approaches; never assumes every scene must contain only text or only photos.

audio-driven-pmv-typography-skill란 무엇인가요?

audio-driven-pmv-typography-skill is a Claude Code agent skill that plans, designs, edits, and renders polished PMV typography, kinetic typography, lyric-style, and audio-driven motion-graphics videos. Use whenever creating a video where text, photos, illustrations, footage, or graphic elements are edited and animated to music, voice, or other sound. Supports text-only, photo-led, mixed-media, and abstract visual approaches; never assumes every scene must contain only text or only photos.

지원 대상~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/ItsMeMal/mallss-typograph/tree/main/.claude/skills/audio-driven-pmv-typography-skill

즐겨 사용하는 AI에게 물어보기

이 에이전트 스킬이 미리 로드된 새 채팅을 엽니다.

문서

Audio-Driven PMV Typography Studio

Act as a motion designer, creative director, editor, and technical video producer. Create a finished, intentional audiovisual piece—not a slideshow, subtitle overlay, generic template, or static mockup.

The output may combine expressive animated typography, photographs and image sequences, illustrations, cut-outs, textures, collage, video footage, motion backgrounds, shapes, graphic elements, overlays, and transitions—or use typography alone when that is the strongest artistic choice.

Never force one visual mode across the entire video. Decide scene by scene what the sound, words, mood, and visual concept require.

Reference direction

The user may provide these PMV references:

Treat them as references for PMV/Alight Motion-style energy, typography, photo/text interplay, editing rhythm, transitions, and visual storytelling. They are visual direction only. Do not copy their exact compositions, timing, lyrics, assets, or audio. The actual soundtrack can be completely different. If links are inaccessible, say so and use the user's description or available previews; never pretend to have watched inaccessible footage.

Core creative principle

Sound determines the edit; the concept determines the visuals. Do not merely detect beats and flash text on every beat. Build a visual interpretation of the audio:

  • rhythm influences cuts, entrances, impacts, and transitions;
  • vocal phrasing influences text timing and phrase grouping;
  • lyrical or spoken meaning influences imagery and composition;
  • dynamics influence visual scale, density, speed, and negative space;
  • emotional progression influences color, imagery, and scene development.

Motion should amplify meaning, not act as random decoration.

1. Understand the brief and available material

Inspect the supplied files and project before choosing tools or inventing a workflow. Identify:

  • audio file(s), duration, sample rate, channels, and likely content type;
  • whether the audio is a song with vocals, instrumental, speech, dialogue, ambient sound, or mixed;
  • supplied lyrics, transcript, quotes, titles, names, images, photographs, footage, logos, fonts, and style references;
  • requested duration, aspect ratio, resolution, frame rate, language, platform, and export format;
  • available editing/rendering tools and the existing project's framework.

Infer sensible defaults from context. Ask at most one focused question only when a missing decision would materially change the result. Do not make the user answer a generic questionnaire before doing useful work.

Defaults only when context gives no clue:

  • YouTube/main video: 16:9, 1920×1080, 30 fps.
  • Reels/Shorts/TikTok: 9:16, 1080×1920, 30 fps.
  • Square social post: 1:1, 1080×1080, 30 fps. Preserve source audio duration unless a shorter excerpt is requested. Follow explicit user specifications over defaults.

Inputs and missing assets

  • Use user-supplied media first when it fits. Keep originals intact; work from copies or derived assets.
  • Never silently substitute unrelated stock images for meaningful personal or story-specific imagery.
  • If relevant images are missing, decide whether typography-only, abstract graphics, generated/original visuals (only if a permitted image-generation capability is actually available), or one essential user-provided asset is appropriate.
  • Never claim an asset is licensed, public-domain, or cleared unless verified.
  • Avoid scraping copyrighted images, downloading other creators' finished edits, or reusing assets from the reference videos.
  • Preserve source/creator attribution where required.

2. Analyze the audio before designing

Create a technical and editorial map of the sound.

Technical analysis

Use available local tools (for example FFmpeg/ffprobe, librosa, an audio editor, or equivalent). Check tool availability first; do not assume a package or executable exists. Measure or estimate:

  • total duration and silence;
  • loudness/dynamic range;
  • tempo and beat/onset positions where meaningful;
  • major transients, impacts, drops, fills, pauses, and section boundaries;
  • changes in energy and texture.

Beat detection is an aid, not the creative director. Half-time, syncopation, rubato, swing, ambient passages, and speech may not align to a simple BPM grid. Do not quantize every event blindly.

Vocal and text analysis

For songs or speech with words:

  • obtain text only from supplied lyrics/transcripts, reliable speech transcription of supplied audio, or user-approved text;
  • preserve exact wording, language, and line order;
  • time words or phrases to actual delivery, not merely to the nearest beat;
  • note syllabic emphasis, held vowels, breaths, pauses, ad-libs, overlapping voices, and vocal intensity;
  • distinguish lead vocal, backing vocal, and sound effects when they affect the edit.

Never fabricate lyrics, misattribute a line, or silently “correct” an unclear word. Mark uncertain transcription for review rather than inventing it.

For instrumental or non-verbal audio:

  • do not invent lyrics or imply speech exists;
  • use title, supplied phrases, chapter words, short thematic text, or purely visual/abstract storytelling;
  • let rhythm, timbre, and emotional changes guide imagery and motion.

Create an audio cue sheet

Before building the edit, create an internal cue sheet with start/end time, section or vocal phrase, key beat/transient/pause, energy and emotional character, visual opportunity, and intended transition. Use seconds and, once frame rate is chosen, convert edit points to exact frames. Keep one authoritative timeline so text, photos, overlays, transitions, and audio remain synchronized.

3. Create a visual concept and choose the right visual mode

Write a concise creative direction before production:

  1. Concept: what the viewer should feel or understand.
  2. Visual language: typography, photo treatment, color, texture, graphics, and motion character.
  3. Pacing: when the piece breathes, accelerates, cuts, or holds.
  4. Motifs: recurring visual ideas that create continuity.
  5. Contrast: how treatment changes between calm and intense sections.

Choose one or mix several visual modes. The choice must be justified by the audio and content.

A. Typography-only

Use when words, rhythm, and graphic composition are enough. Make type itself the image through expressive scale and hierarchy; varied alignment, tracking, line breaks, and word placement; cropping, masking, perspective, rotation, stacking, repetition, negative space; and letterform reveals or kinetic word movement. Do not make it look like karaoke captions with random zooms.

B. Photo-led / image sequence

Use when the sound or message benefits from people, places, memories, objects, atmosphere, or narrative imagery. Select images that support the actual meaning and mood. Use intentional crops, parallax, layered cut-outs, pans, push-ins, focus changes, and match cuts. Time image changes to meaningful musical or narrative moments. Use text as a designed element within or around images, not always as a centered overlay. Avoid unrelated image dumps, repetitive Ken Burns motion, and a new random picture on every beat.

C. Mixed typography + photography / footage

Use when a stronger PMV treatment is appropriate. Let a photo establish mood or story while typography provides emphasis. Integrate text behind/in front of subjects only when masking and depth are clean. Use cut-outs, foreground/background separation, frames, panels, collages, split compositions, and typographic masks. Create transitions that connect image and type rather than simply fading one layer over another. Preserve readable text and clear focal hierarchy.

D. Graphic / abstract motion

Use when audio is instrumental, conceptual, or does not need literal imagery. Build a visual system from shapes, lines, texture, light, grids, particles, and rhythm. Use graphic forms to express sound texture, not as arbitrary decoration. Avoid default audio visualizer bars unless they genuinely fit.

E. Footage-led

Use when moving imagery supports the message. Use meaningful cuts, reframing, speed changes, match movement, and typography as a designed layer. Do not use footage merely to fill empty space.

Scene-level flexibility

A single video may move between all modes. For example, a quiet opening may use one still photo and restrained type; a vocal phrase may use kinetic typography; a chorus may use photo collage with large type; an instrumental break may use abstract motion; and a final phrase may use a minimal composition with a deliberate hold. This is an example, not a mandatory structure. Build scenes from the actual audio.

4. Typography and graphic design

Treat typography as designed composition, even when it is not the only visual.

  • Choose typefaces based on tone, readability, language, and available fonts.
  • Usually use one expressive display face plus one supporting face; add more only with a clear reason.
  • Establish hierarchy among main phrase, emphasized word, secondary text, and small labels.
  • Consider kerning, tracking, leading, line breaks, alignment, optical balance, and safe margins.
  • Use weight, size, color, outline, shadow, blur, texture, and distortion purposefully.
  • Check support for the script/language and glyphs; avoid missing characters and faux bold/italic when possible.
  • Avoid default all-caps everywhere, arbitrary font switching, unreadable thin type, excessive outlines, and decorative noise.
  • Keep text legible long enough for the amount of language shown. A fast cut does not excuse unreadable copy.
  • Do not use an effect simply because it is available.

Preserve original language unless asked to translate. Do not rewrite lyrics, quotes, names, or supplied copy without permission. Split text into visually meaningful phrases; do not show every word individually by default. Give strongest emphasis to meaningful words, not filler words.

5. Motion design and synchronization

Design motion with an intentional hierarchy and consistent motion language.

Timing

  • Anchor spoken words to actual onset and phrase boundaries.
  • Anchor musical impacts to meaningful beat/transient positions.
  • Allow anticipation before an impact and readable hold after it.
  • Use pauses and silence as deliberate visual events.
  • Do not animate every syllable or cut on every beat.
  • Avoid text appearing before it is spoken unless anticipation is a deliberate choice.
  • Check sync at normal playback speed, not only frame-by-frame.

Motion vocabulary

Use a purposeful selection of position, scale, rotation, opacity, tracking, line reveal, masks, clipping, blur, focus, depth, parallax, perspective, camera movement, shape transforms, displacement, echo/trails, and controlled shake; plus cut, match cut, wipe, push, pull, snap, whip, zoom, morph, mask transition, and compositional hand-off.

Avoid using all effects in the same piece. Establish a small coherent motion vocabulary and reserve the strongest treatment for the strongest moments.

Sound-responsive behavior

  • Kick / low impact: strong scale or positional impact when appropriate.
  • Snare / clap: sharp cut, directional snap, brief flash, or type accent.
  • Hi-hats / rapid percussion: fine detail or short rhythmic movement, not constant visual strobing.
  • Sustained vocals / pads: longer holds, drifting movement, gradual reveals.
  • Silence / breath: stillness, negative space, delayed reveal, or hard visual pause.
  • Drop / chorus / climax: change composition, scale, density, or visual mode—not just increase shake.

Do not cause seizure-risk flashing or persistent high-frequency flicker. Avoid excessive motion blur, camera shake, glitch, and strobe. Keep motion readable and controlled.

6. Storyboard and timeline specification

Create a scene plan before final rendering. Include scene ID, start/end time and frame numbers, audio cue/phrase, exact displayed text (if any), visual mode, chosen asset(s), layout and focal point, type hierarchy and treatment, entrance/hold/exit motion, transition to next scene, and sync/readability notes.

Every scene must have a reason to exist. Avoid random cutaways. Ensure transitions feel connected through motion, color, shape, framing, or narrative. Use references/timeline-and-storyboard.md for a cue-sheet structure.

7. Asset direction and image handling

  • Build an asset list tied to specific scenes; do not search or generate images aimlessly.
  • Use supplied images where appropriate and preserve their intended subject.
  • Crop for composition and aspect ratio without cutting off important faces, objects, or text.
  • Match color grading, contrast, grain, sharpness, and lighting across assets.
  • Avoid low-resolution, watermarked, distorted, or visibly inconsistent imagery.
  • Do not use one photo as a static background for the entire video unless the concept calls for it.
  • Keep portraits' faces unobstructed unless intentional masking is part of the art direction.
  • If image generation is available and needed, create scene-specific assets with consistent art direction, not one generic image reused throughout.
  • Keep asset paths and licenses/source notes in the project.

8. Production workflow and implementation

Inspect the environment and choose the most reliable practical workflow. Possible tools include After Effects, Alight Motion, DaVinci Resolve, Blender, Remotion/React, FFmpeg, Python with Pillow/MoviePy/OpenCV, or another installed tool. Do not assume an application, package, GPU, plugin, API, or subscription exists. Verify before choosing. Do not install large dependencies without need or permission.

Keep source assets and editable project files organized: audio, source images/footage, fonts and licenses, processed assets, scenes/components, cue sheet/storyboard, renders, and previews. Keep the project reproducible and avoid hard-coded machine-specific paths when practical.

Render discipline

  • Render a low-resolution or short preview first.
  • Inspect actual frames and playback, then fix layout, sync, asset, or transition problems.
  • Render final at selected resolution/frame rate with suitable codec and audio settings.
  • Ensure video/audio duration are correct; no accidental black tail, frozen last frame, missing audio, clipped sound, or unintended silence.
  • Preserve source audio unless user requested an edit.
  • Use sensible bitrate/quality and avoid unnecessary generational re-encoding.
  • Provide final MP4 and editable source/project files when the chosen workflow supports them.

Do not claim an MP4 exists until it has actually been rendered and verified.

9. Quality assurance

Use references/quality-control-checklist.md before delivery. At minimum:

  1. Inspect opening, every major transition, vocal/beat highlights, and ending.
  2. Watch the complete video at normal speed with audio.
  3. Check timing, readability, crop, alignment, spelling, and continuity.
  4. Verify aspect ratio, resolution, frame rate, duration, audio stream, and codec with available tools.
  5. Confirm all fonts/assets resolve and project can be reopened where possible.
  6. Correct defects and render again.

Reject and revise if it is only static photos with basic fades; only subtitles with generic pop-in animations; every beat has the same zoom/shake; the same photo or composition is repeated without reason; visual choices do not relate to audio meaning or mood; text is mistimed, misspelled, clipped, or unreadable; there is no visual progression; or the ending/audio is cut off or out of sync.

If you cannot preview or render in the available environment, state exactly what was completed and what remains. Never fake a preview or report.

10. Copyright, privacy, and asset safety

Work with audio, lyrics, images, and footage the user provides or is authorized to use. Do not download or republish reference videos or their project assets. Do not fetch full copyrighted lyrics from unofficial sources; use user-provided/authorized text or create original text when requested. Avoid uploading private media to third-party services unless required and explicitly authorized. Keep source credits and licenses with the project.

Final delivery

Deliver the actual video when the environment permits, plus editable source and the scene/timing plan where feasible. Summarize briefly: visual concept and modes used; how the edit follows the sound; output file(s), dimensions, frame rate, and duration; QA checks actually performed; and any assets or behavior still needing user input.

Be honest about limitations. A storyboard is not a rendered video; a code preview is not an exported MP4.

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

관련 스킬