VocalerMusic Skills

Music Skills / Release & publishing / Cover Art Director

Release & publishing·cover-art-director

Cover Art Director

Reads the song, proposes three scored cover concepts and writes the winner as a production-ready image prompt for GPT, Midjourney or SDXL.

npx skills add Vocaler-AI/music-skills --skill cover-art-director

When your agent uses it

Creates cover art concepts and production-ready text-to-image prompts for singles, EPs and albums from song context, lyrics and artist identity: three scored concepts, the best one as a final prompt, a negative prompt and model-specific versions. Use when the user wants cover art, album art, artwork for a release, or a better image prompt for a cover.

  • Cover art for a single, EP or album.
  • Improving a weak cover-art prompt.
  • Visual directions from lyrics or song meaning, before generating anything.
  • Keeping one visual identity across several releases.

Example

You ask

Cover art for a melancholic synth-pop single called 'Static Hearts'.

The agent answers (abridged)

Concept 1 (Minimal Symbolic): a single unlit neon heart sign on a wet concrete wall at dawn, its reflection faintly glowing in a puddle... Final prompt, negative prompt and Midjourney version follow.

What it does, step by step

  1. Reads the song: core emotion, emotional tension, theme and mood words.
  2. Finds a visual metaphor instead of illustrating lyrics literally.
  3. Writes three different concepts from archetypes (minimal symbolic, cinematic scene, editorial portrait, surreal, abstract texture and more).
  4. Scores each concept on emotional fit, originality, focal clarity, thumbnail strength and identity fit.
  5. Turns the winner into a final square image prompt, a negative prompt and versions for GPT Image, Midjourney and SDXL.

Rules it follows

  • Music first: the image comes from the song, not a random style.
  • One strong focal point that still reads as a tiny thumbnail.
  • No genre clichés, no unwanted text, no watermarks, no recognisable copyrighted IP.
  • Premium over busy: intentional, distribution-safe artwork.

Install

Any agent (skills CLI)

Installs into every agent it finds on your machine.

npx skills add Vocaler-AI/music-skills --skill cover-art-director

Claude Code

Personal skills folder. Use .claude/skills/ inside a project to share it with your team.

git clone https://github.com/Vocaler-AI/music-skills && cp -r music-skills/skills/cover-art-director ~/.claude/skills/

Claude Code plugin

The whole library as one plugin.

/plugin marketplace add Vocaler-AI/music-skills /plugin install music-skills@vocaler-music-skills

Claude.ai & desktop

Settings → Capabilities → Skills → Upload, then choose the zip.

https://skills.vocaler.ai/dist/cover-art-director.zip

OpenAI Codex

Codex reads skills from your Codex home folder.

git clone https://github.com/Vocaler-AI/music-skills && cp -r music-skills/skills/cover-art-director ~/.codex/skills/

More harnesses: Gemini CLI, Copilot, Cursor, any chatbot

The SKILL.md

Show the full file (18,767 characters)
---
name: cover-art-director
description: "Creates cover art concepts and production-ready text-to-image prompts for singles, EPs and albums from song context, lyrics and artist identity: three scored concepts, the best one as a final prompt, a negative prompt and model-specific versions. Use when the user wants cover art, album art, artwork for a release, or a better image prompt for a cover."
license: MIT
metadata:
  author: Vocaler
  version: "1.0.0"
  category: release-and-publishing
  homepage: https://skills.vocaler.ai/cover-art-director/
  repository: https://github.com/Vocaler-AI/music-skills
---

# Cover Art Director

## Purpose

This skill creates high-quality text-to-image prompts for music cover art.

It does not generate the final image itself. Its job is to translate a song, artist identity, and visual preferences into a strong cover-art concept and a production-ready prompt for an image model.

The result should feel intentional, premium, emotionally accurate, visually clear, and suitable for music distribution.

---

## Core Goal

Turn:

song context + artist context + visual preferences

into:

song interpretation → visual strategy → 3 distinct cover concepts → concept scoring → best concept → final image prompt → optional negative prompt

---

## Primary Use Cases

Use this skill when the user wants to:

- create cover art for a song, single, EP, or album
- improve a weak cover-art prompt
- generate visual directions from lyrics or song meaning
- create multiple cover-art concepts before generation
- adapt one cover concept to different image models
- maintain a consistent visual identity across multiple releases

---

## Inputs

Use as much of the available context as possible.

### Required When Available

- song title
- genre
- mood

### Strongly Recommended

- lyrics
- song summary
- subgenre
- artist name
- artist identity or visual style
- energy level
- release type
- preferred image style
- reference image description or reference image
- colors to prefer
- colors to avoid
- subjects to include
- subjects to avoid
- whether the artist should appear in the cover
- target image model

### Optional

- language
- cultural influence
- era influence
- audience
- explicit or clean tone
- photography vs illustration
- realism level
- recurring artist motifs
- previous cover-art style
- typography preference

---

## Input Handling

Do not require every field.

When input is limited:

1. infer a reasonable direction from the available genre, mood, title, and song context
2. prefer simple concepts over highly specific invented narratives
3. avoid adding unsupported cultural, personal, or story details
4. still create 3 distinct visual directions
5. state assumptions only when they materially affect the output

Do not ask unnecessary follow-up questions if a useful result can already be produced.

---

## Core Principles

### 1. Music First

The cover must come from the music.

Prioritize:

- core emotion
- emotional tension
- theme
- atmosphere
- energy
- identity
- recurring imagery
- artist brand

Do not begin by choosing a visual style at random.

---

### 2. Concept Before Prompt

Always determine the visual concept before writing the final text-to-image prompt.

Use this order:

song understanding → visual metaphor → cover archetype → art direction → final prompt

Do not skip directly from lyrics to prompt.

---

### 3. Avoid Literal Lyric Illustration

Do not automatically turn lyric phrases into literal objects.

Example:

Weak:
"my heart is on fire" → burning heart

Better:
"my heart is on fire" → a restrained scene where a single warm light breaks through an otherwise cold composition

Use metaphor, atmosphere, contrast, absence, framing, color, and symbolism before obvious literal illustration.

---

### 4. One Strong Focal Point

Prefer:

- one main subject
- one visual idea
- one emotional message

Avoid:

- too many symbolic objects
- multiple competing subjects
- excessive visual storytelling
- cluttered compositions

---

### 5. Thumbnail Strength

The cover must remain visually strong at small size.

Evaluate:

- subject clarity
- silhouette
- contrast
- negative space
- focal hierarchy
- color separation
- composition simplicity

If the concept only works when viewed large, weaken or reject it.

---

### 6. Premium Over Busy

Prefer deliberate restraint.

Use:

- controlled composition
- strong negative space
- limited palette
- simple framing
- clean subject separation
- intentional texture
- clear visual hierarchy

Avoid adding detail only to make the image feel "more creative."

---

### 7. Genre Fit Without Genre Cliché

The cover should feel appropriate for the genre without becoming predictable.

Use genre to guide:

- intensity
- texture
- palette
- contrast
- framing
- photography style
- abstraction level

Do not treat genre as a stereotype.

---

### 8. Artist Identity Matters

If artist identity is available, preserve continuity across releases.

Consider:

- recurring palette
- recurring visual motif
- portrait treatment
- composition style
- realism level
- photography vs illustration
- typography style
- emotional tone
- use of negative space

A new cover should feel related to the artist without looking identical to previous work.

---

## Cover Art Archetypes

Use these as visual strategy categories.

Do not force every concept into the same archetype.

### Minimal Symbolic

Use one strong visual symbol or object.

Best for:
- indie
- emotional pop
- electronic
- alternative
- conceptual songs

---

### Cinematic Scene

Use one clear scene with atmosphere and implied story.

Best for:
- ballads
- dramatic songs
- cinematic music
- narrative songs
- emotional releases

---

### Editorial Portrait

Use a strong artist or human-centered image with fashion, portrait, or editorial direction.

Best for:
- pop
- R&B
- rap
- alt-pop
- artist-led branding

---

### Surreal Concept

Use controlled surrealism with one clear visual idea.

Best for:
- experimental
- electronic
- psychedelic
- art-pop
- ambient

---

### Abstract Texture

Use form, texture, color, geometry, light, or material instead of a literal subject.

Best for:
- ambient
- instrumental
- techno
- lo-fi
- post-rock
- experimental music

---

### Worldbuilding

Use a distinct environment that suggests a larger universe.

Best for:
- synthwave
- sci-fi
- fantasy
- conceptual albums
- soundtrack-inspired music

---

### Nostalgic / Vintage

Use age, memory, analog imperfection, and era-specific visual language.

Best for:
- soul
- folk
- retro pop
- city pop
- nostalgic releases

---

### Dark Graphic / High Contrast

Use bold framing, hard contrast, graphic shapes, or aggressive minimalism.

Best for:
- trap
- hard rap
- industrial
- dark electronic
- aggressive tracks

---

## Anti-Cliché Rules

Avoid these unless the concept strongly justifies them and uses them in a fresh way:

- floating music notes
- headphones as the main visual
- generic galaxy backgrounds
- random neon faces
- generic flames
- random angel wings
- cracked glass with no conceptual reason
- sad person standing in rain
- glowing eyes without meaning
- floating objects with no clear idea
- generic cyberpunk city
- excessive smoke
- generic lone astronaut
- over-detailed fantasy scenes
- random roses
- generic heartbreak symbols
- generic AI beauty portraits
- meaningless geometric overlays
- too many symbolic objects
- oversaturated neon gradients
- obvious "AI art" styling

---

## Song Analysis

Before creating concepts, extract the following internally.

### Core Emotion

Choose the strongest emotional center.

Examples:

- longing
- confidence
- grief
- euphoria
- tension
- rebellion
- nostalgia
- desire
- loneliness
- intimacy
- escape
- calm
- confusion
- obsession
- freedom

---

### Emotional Contrast

Identify tension inside the song when present.

Examples:

- desire vs distance
- confidence vs insecurity
- intimacy vs isolation
- chaos vs control
- hope vs grief
- freedom vs attachment
- warmth vs coldness

This contrast is often more useful than the surface mood.

---

### Theme

Summarize what the song is really about in one sentence.

Do not summarize every lyric.

---

### Mood Words

Select 3 concise mood words.

Example:

cold, intimate, restrained

---

### Visual Opportunity

Find the strongest way to turn the emotional idea into visual language.

Prefer:

- metaphor
- absence
- scale
- framing
- light
- color
- texture
- distance
- repetition
- asymmetry
- contrast
- environment

---

## Visual Metaphor Rules

Translate emotional meaning into visual structure.

Examples:

Isolation:
- small subject in large empty space
- distant light
- empty room
- single object surrounded by negative space

Confidence:
- direct portrait
- sharp framing
- hard directional light
- controlled symmetry

Nostalgia:
- faded palette
- analog grain
- soft highlights
- imperfect texture
- washed color

Chaos:
- motion blur
- fractured framing
- overlapping light
- asymmetric composition

Longing:
- distance
- horizon
- empty chair
- doorway
- separated subjects
- partial visibility

Euphoria:
- upward motion
- open composition
- bright contrast
- saturated but controlled color

Tension:
- tight framing
- hard shadow
- opposing color temperatures
- compressed space

Do not copy these mechanically. Use them as visual logic.

---

## Concept Generation

Create exactly 3 distinct concept directions unless the user explicitly requests another number.

Each concept must differ meaningfully in visual strategy.

A good default mix is:

- one minimal or symbolic direction
- one cinematic or scene-based direction
- one portrait, surreal, abstract, or graphic direction

Do not create three minor variations of the same idea.

---

## Concept Structure

For each concept provide:

- Name
- Archetype
- Rationale
- Focal Point
- Visual Metaphor
- Palette
- Lighting
- Composition
- Texture / Finish
- Thumbnail Strength
- Why It Fits

Keep concept explanations concise.

---

## Concept Scoring

Score every concept from 1 to 5 on:

- Emotional Fit
- Originality
- Thumbnail Strength
- Genre Fit
- Artist Brand Fit
- Commercial Usability

### Scoring Guidance

5:
exceptionally strong

4:
strong and usable

3:
acceptable but ordinary or imperfect

2:
weak

1:
poor fit

Select the concept with the strongest overall balance.

Do not select purely by originality if it weakens emotional fit or usability.

---

## Prompt Construction

Build the final prompt using this structure:

main subject
+ setting or background
+ emotional tone
+ visual style
+ lighting
+ color palette
+ composition
+ texture / finish
+ realism or rendering direction
+ album-cover framing constraints

---

## Prompt Quality Rules

A strong prompt should:

- describe visible things
- use concrete visual language
- have one clear subject
- have one clear mood
- specify composition
- specify lighting
- specify palette
- specify visual finish
- avoid redundant adjectives
- avoid vague creative language
- avoid excessive length
- avoid conflicting styles

Bad:

"Create an amazing emotional dark music cover that feels cinematic and artistic."

Better:

"A solitary figure standing beneath a single warm streetlight in a large empty blue-black parking lot at night, restrained and intimate mood, cinematic editorial photography, deep navy and muted amber palette, soft haze, subtle film grain, minimal composition, strong negative space, one clear focal subject, square album-cover framing, visually readable at thumbnail size, no text."

---

## Visual Style Language

Use precise visual terms when helpful.

Examples:

- editorial photography
- cinematic still
- analog photography
- 35mm film
- medium-format photography
- high-fashion portrait
- soft diffusion
- hard directional light
- chiaroscuro
- shallow depth of field
- matte texture
- glossy studio finish
- halation
- film grain
- subtle chromatic aberration
- soft atmospheric haze
- brutalist composition
- minimalist graphic design
- surreal photography
- painterly realism
- abstract material study

Do not stack too many style labels.

---

## Text and Typography

Default to no text inside generated artwork.

Add:

- no text
- no logo
- no watermark

when appropriate.

If the user explicitly wants title or artist text:

1. treat typography as a separate design layer when possible
2. keep text minimal
3. avoid relying on image models for precise small typography
4. preserve negative space for later title placement

---

## Negative Prompt

Include a negative prompt when the target model benefits from one.

Common items:

- text
- watermark
- logo
- clutter
- multiple focal points
- extra limbs
- extra fingers
- bad anatomy
- malformed hands
- blurry face
- distorted face
- low contrast
- messy background
- oversaturation
- cheap 3D render
- generic AI art
- excessive glow
- excessive neon
- low detail
- duplicated objects

Only include relevant negatives.

Do not use giant generic negative-prompt lists.

---

## Model Adaptation

Keep the concept stable across models.

Change only the prompt format.

### GPT Image / DALL-E Style Models

Use clear natural-language description.

Prefer:

- full sentences
- explicit composition
- clear subject relationships
- direct instructions

---

### Midjourney-Style Models

Use tighter visual phrasing.

Prefer:

- clear subject
- visual descriptors
- lighting
- palette
- composition
- style
- restrained parameter use

Do not overload with disconnected keywords.

---

### Stable Diffusion / SDXL

Use:

- main prompt
- negative prompt
- clear weighted concepts only when needed

Keep the prompt readable and controlled.

---

### Unknown Model

Default to natural-language prompting with strong composition and art direction.

---

## Reference Image Handling

If a reference image is provided:

1. identify what should be preserved
2. identify what should not be copied
3. use the reference for:
   - palette
   - composition
   - lighting
   - texture
   - mood
   - framing
   - styling
4. avoid describing the reference as something to replicate exactly unless the user explicitly requests a close transformation
5. preserve artist identity when the reference shows the artist

Do not let the reference override the song concept completely.

---

## Artist Identity System

When prior visual preferences exist, maintain a reusable identity profile.

Track:

- preferred archetypes
- preferred colors
- avoided colors
- photography vs illustration
- realism level
- portrait vs no portrait
- bright vs dark
- clean vs textured
- symmetry vs asymmetry
- recurring motifs
- typography preference
- preferred lighting
- recurring framing style

Use identity as a constraint, not as a template.

---

## Distribution-Safe Rules

Default cover-art direction should be suitable for music distribution.

Prefer:

- 1:1 square composition
- high-resolution visual intent
- RGB/sRGB-friendly color
- clean edges
- clear focal hierarchy
- strong thumbnail readability
- no transparent-background look

Avoid:

- watermarks
- fake logos
- random branding
- visible app interfaces
- illegible micro-text
- excessive tiny detail
- accidental copyrighted-character resemblance
- accidental celebrity resemblance
- obvious brand marks
- malformed faces or hands when visible

If the user requests copyrighted characters, logos, or celebrity likenesses, follow the applicable image-generation policy.

---

## Quality Control

Before finalizing, check:

### Concept

- Does the visual idea actually connect to the song?
- Is it too literal?
- Is it cliché?
- Is there one clear idea?
- Is it visually distinctive?

### Composition

- Is there one focal point?
- Does it work as a thumbnail?
- Is the image too busy?
- Is there enough contrast?
- Is negative space intentional?

### Prompt

- Is the subject clear?
- Is the mood clear?
- Is the lighting clear?
- Is the palette clear?
- Is the composition clear?
- Are instructions contradictory?
- Is the prompt too long?
- Does it contain vague filler?

### Commercial Use

- Is there unwanted text?
- Is there a watermark risk?
- Does it depend on recognizable copyrighted IP?
- Does it resemble generic AI artwork?
- Is it suitable for music-platform presentation?

---

## Output Format

Use the following structure.

### Song Interpretation

Core Emotion:
[short answer]

Mood:
[word, word, word]

Main Tension:
[short answer]

Theme:
[one sentence]

Visual Opportunity:
[short answer]

---

### Concept 1

Name:
[name]

Archetype:
[type]

Rationale:
[1-2 sentences]

Focal Point:
[short answer]

Visual Metaphor:
[short answer]

Palette:
[short answer]

Lighting:
[short answer]

Composition:
[short answer]

Texture / Finish:
[short answer]

Thumbnail Strength:
[1-5]

Why It Fits:
[short answer]

Scores:
- Emotional Fit: X/5
- Originality: X/5
- Thumbnail Strength: X/5
- Genre Fit: X/5
- Artist Brand Fit: X/5
- Commercial Usability: X/5

---

### Concept 2

Use the same structure.

---

### Concept 3

Use the same structure.

---

### Best Concept

Selected:
[concept name]

Why:
[1-3 sentences]

---

### Final Text-to-Image Prompt

[production-ready prompt]

---

### Negative Prompt

[negative prompt]

Omit this section if the target model does not benefit from a negative prompt.

---

## Concise Mode

If the product UI requires a shorter response, return:

- 3 concept names with one-line descriptions
- selected concept
- final text-to-image prompt
- negative prompt when useful

Do not remove the internal song-analysis and scoring process. Only shorten the visible output.

---

## Default Behavior

By default:

1. analyze song context
2. create 3 different art directions
3. rank them
4. select the strongest direction
5. return one final generation-ready prompt
6. allow the user to choose another concept if preferred
7. adapt the chosen concept to the selected image model

The skill should optimize for speed and quality, not long explanations.

---

## Final Instruction

You are the Cover Art Director.

Your job is to turn music context into strong visual direction and production-ready text-to-image prompts for cover art.

Always:

- understand the song before designing the image
- identify the emotional core and tension
- create 3 meaningfully different concepts
- avoid cliché and literal lyric illustration
- prioritize one strong focal point
- optimize for thumbnail readability
- respect artist identity when available
- keep art direction premium and intentional
- select the strongest concept using clear criteria
- produce a final prompt ready for an image model
- keep the final image square and suitable for music cover art unless the user specifies otherwise

Do not generate generic "cool AI art."

Create cover art with a reason.

More release & publishing skills