Art Direct

Art Direct

Turn content into visual direction. Point this at anything — a deck, a document, an essay, a brief, a webpage — and get back a creative direction you can actually execute.

When to Use

• User shares a file (any format) and needs visuals for it
• Developing visual identity for content before building/designing
• Translating written material into photography/illustration direction
• Creating image prompts for AI generation tools
• Art directing a presentation, document, report, or website
• Reviewing existing visuals against content intent (critique mode)

Supported Inputs

Read content from whatever the user provides:

If a format doesn't extract cleanly, ask the user to paste the text.

The Workflow

INGEST CONTENT → ANALYSE → PROPOSE 2-3 DIRECTIONS → USER SELECTS → VISUAL BRIEF + PROMPTS

Stage 1: Content Ingestion & Analysis

Read the full content. Extract:

1. Structure — What are the units? (slides, sections, chapters, paragraphs, pages)
2. Core themes — The 2-3 big ideas the content is actually about
3. Narrative arc — Does it build? Contrast? Layer? List?
4. Audience — Who receives this? What do they expect to see?
5. Tone — Authoritative? Inspirational? Intimate? Provocative? Technical?
6. Key moments — Which sections carry the most weight, demand the strongest visuals?
7. Existing visual language — If the content already has images, assess what's working and what isn't

Output a brief content summary before proceeding. Keep it tight — this is for alignment, not a book report.

Stage 2: Creative Direction Proposals

If house style template provided (--style <name>):

• Validate content fits the style
• Note any tensions and how to bridge them
• Skip to Stage 3 with adapted style guide

If no house style:

Propose 2-3 distinct visual directions. Each must be genuinely different — not three shades of the same idea. For each:

DIRECTION: [Name — a short handle like "Archival Authority" or "Warm Machinery"]

MOOD
What it feels like: [emotional quality in 2-3 words]
Energy: [calm / dynamic / tense / contemplative / electric]

PHOTOGRAPHY STYLE
Type: [documentary / editorial / conceptual / abstract / archival / illustrative]
Subjects: [what appears in the images]
Lighting: [quality of light]
Color treatment: [warm/cool shift, saturation, film stock reference]
Composition: [framing approach]

REFERENCE TOUCHSTONES
"Think [X] meets [Y]" — cite real publications, campaigns, photographers, or brands

WHAT THIS DIRECTION AVOIDS
[Specific clichés and visual tropes this direction rejects]

WHY THIS FITS THE CONTENT
[1-2 sentences connecting direction to content themes]

Present all directions. User picks one (or asks for a hybrid). Lock the choice.

Stage 3: Visual Style Guide

Once direction is selected, output the working style guide:

VISUAL STYLE GUIDE: [Content Title]
Direction: [Chosen direction]

PHOTOGRAPHY STYLE
─────────────────
Type: [Documentary / Editorial / Conceptual / Abstract / Archival]
Subjects: [What to feature — specific, not generic]
Composition: [Framing rules]
Lighting: [Light quality]
Color treatment: [Color approach, film stock if relevant]

MOOD & TONE
───────────
Primary emotion: [e.g., quiet confidence]
Supporting emotions: [e.g., warmth, precision]
Energy level: [Calm / Dynamic / Tense / Contemplative]

CONSISTENCY RULES
─────────────────
• [Shared quality all images must have]
• [Human subject guidelines]
• [Color palette anchors — hex codes]
• [Aspect ratio defaults]

CLICHÉ BLACKLIST
────────────────
• [Content-specific images to reject]
• [Generic tropes to avoid]
• [Overused metaphors for these themes]

AI GENERATION DEFAULTS
──────────────────────
Photography suffix: [standard prompt additions for photo-style generation]
Illustration suffix: [standard prompt additions for illustration-style generation]

Stage 4: Section-by-Section Execution

Work through the content in its natural units (slides, sections, chapters, key passages). For each:

Step 1: Interpret the section's job

What must the visual communicate? What's the emotional beat?

Step 2: Apply the Five-Lens Framework

Generate options through five lenses:

Step 3: Output the Visual Brief

For the recommended lens (guided by the style guide's lens preferences), output:

SECTION: "[Section title or key line]"
VISUAL JOB: [What this image must do]
LENS: [Which lens and why]

CONCEPT
[2-3 sentence description of the exact image — specific enough
that a photographer could shoot it or a designer could find it]

AI GENERATION PROMPTS
─────────────────────
MIDJOURNEY:
[Full prompt with style suffixes, --ar, --v, --style flags]

DALL-E / GPT IMAGE:
[Natural language prompt optimized for DALL-E]

GEMINI:
[Prompt formatted for Gemini image generation]

IDEOGRAM:
[Prompt formatted for Ideogram, especially for any text-in-image needs]

SOURCING GUIDANCE
─────────────────
If searching (not generating):
  Search: [2-3 specific, refined search queries]
  Where: [Specific sources — see Source Guide below]
  Avoid: [What will come up that you should skip]

ALTERNATIVES
────────────
[1-2 other lens options briefly described, in case the primary doesn't land]

Step 4: For content with many sections

Don't generate all sections unprompted. Output:

1. The first 2-3 sections as examples
2. A summary table of all remaining sections with recommended lens and one-line concept
3. Ask which sections to develop fully

The Five-Lens Framework (Detail)

For any concept, five ways to see it:

The oblique lens is the hardest and the most valuable. It's the image that makes someone stop and think. Use it for hero images and opening sections.

Source Guide

Do not default to stock photo sites. Stock search produces generic results regardless of how specific your terms are. Instead:

Primary: AI Generation

The best match for precise creative vision. Generate exactly what the concept describes.

• Midjourney — Best for photographic realism and cinematic quality
• DALL-E / GPT Image — Best for conceptual and illustrative work
• Gemini — Good for diagrams, text-in-image, data visualization
• Ideogram — Best when image includes readable text or typography

Secondary: Editorial & Archival Sources

When you need real photography (historical, documentary, journalistic):

• Getty Editorial — Photojournalism, historical archives
• Magnum Photos — Documentary photography
• Library of Congress — US historical archives, public domain
• NASA Image Gallery — Space, earth science, technology
• Wikimedia Commons — Public domain, historical
• British Museum / Smithsonian — Historical objects and documents
• Internet Archive — Historical documents, publications, ephemera
• Google Arts & Culture — Museum collections, artworks

Tertiary: Curated Stock (when you must)

• Unsplash — Best for environmental/atmospheric shots, not people
• Pexels — Acceptable for textures, backgrounds, abstract
• Avoid for: People, business scenarios, technology in use, anything conceptual

For Specific Needs

Anti-Cliché Guide

Universal Blacklist

These images are invisible — viewers have seen them thousands of times:

• Handshakes (any kind)
• Lightbulb = idea
• Puzzle pieces connecting
• Person on mountain summit
• Hands holding globe
• Diverse team pointing at whiteboard
• Plant sprouting = growth
• Rocket = launch/speed
• Chess = strategy
• Maze = complexity
• Road diverging in forest = choice
• Iceberg = hidden depth
• Bridge = connection

The Reframing Technique

When you catch yourself reaching for a cliché:

1. Name the cliché — "I'm about to search for a lightbulb"
2. Ask: What does this concept feel like? — Not look like. Feel like.
3. Ask: What moment captures this for a real person? — Specificity kills cliché
4. Generate that instead

Example: "Innovation"

• Cliché: Lightbulb, circuit board, rocket
• Feels like: The moment before you know if it works
• Real moment: Engineer's hand hovering over a switch, not yet thrown
• That's the image

Critique Mode

When pointed at content that already has images (existing deck, webpage, document):

Step 1: Ingest & View Everything

Read all content. View every image. Do the work before speaking.