prompt-mastery

Prompt Mastery for AI Image Generation

Transform casual image descriptions into professional, high-quality prompts using patterns extracted from 1,186 viral AI-generated images.

The 6 Core Optimization Rules

Rule 1: Professional Terms Over Feeling Words

Replace vague aesthetic words with specific professional terminology, proper nouns, brand names, or artist names.

Instead of...	Use...
Cinematic, vintage	Wong Kar-wai aesthetics, Saul Leiter style
Film look, retro	Kodak Vision3 500T, Cinestill 800T
Warm tones	Sakura Pink, Golden Hour warmth
Japanese style	Wabi-sabi aesthetics, MUJI visual language
High-end design	Swiss International Style, Bauhaus functionalism

Key terminology banks:

• Photographers: Annie Leibovitz, Christopher Doyle, Saul Leiter
• Film stocks: Kodak Vision3 500T, Cinestill 800T, Fujifilm Superia
• Aesthetics: Wabi-sabi, Bauhaus, Swiss International Style

Rule 2: Quantified Parameters Over Adjectives

Replace subjective adjectives with specific technical parameters.

Instead of...	Use...
Professional looking	90mm lens, f/1.8, high dynamic range
From above	45-degree overhead angle
Soft lighting	Soft side backlight, diffused light
Blurred background	Shallow depth of field, f/1.4 bokeh
Dramatic	Volumetric light, chiaroscuro lighting
Wide shot	16mm wide-angle lens

Rule 3: Negative Constraints

Explicitly state what NOT to include to prevent unwanted elements.

Common constraints:

• No text or words allowed
• No watermarks or logos
• No low-key dark lighting (unless intended)
• No high-saturation neon colors
• Product must not be distorted or warped
• Maintain realistic facial features

Rule 4: Sensory Stacking

Go beyond visual descriptions by adding multiple sensory dimensions.

Sensory layers:

• Visual: Color, light, shadow, composition (baseline)
• Tactile: "Texture feels tangible", "Soft velvet surface"
• Olfactory: "Aroma penetrates the frame", "Fresh morning dew scent"
• Motion: "Steam wisps rising", "Fabric gently flowing"
• Temperature: "Steamy warmth", "Crisp cold air"

Rule 5: Group and Cluster

For complex scenes, organize information into logical groups.

Standard grouping pattern:

[Subject Description]

Visual Style:
[Aesthetic references, color palette, artistic style]

Lighting & Atmosphere:
[Light sources, mood, environmental conditions]

Technical Parameters:
[Lens, aperture, film stock, resolution]

Constraints:
[What to avoid, what to preserve]

Rule 6: Format Adaptation

Choose format based on complexity:

• Simple scenes (single subject): Natural language paragraphs
• Complex scenes (multiple elements): Structured groupings with headers

Scene-Specific Patterns

Consult references/scene-guide.md for detailed patterns by scene type:

• Product Photography
• Portrait Photography
• Food Photography
• Cinematic Scenes
• Japanese Aesthetic
• Design/Poster

Smart Genre-Based Optimization

When generating an image, follow this intelligent flow:

Step 1: Auto-Detect Genre

Analyze the user's request and classify into one of these genres:

Genre	Trigger Keywords
food	dish, meal, cuisine, recipe, ingredient, cooking
portrait	person, face, woman, man, selfie, headshot
product	product, packaging, brand, commercial, advertising
3d	icon, render, 3D, isometric, emoji, character
cinematic	movie, scene, dramatic, action, poster
design	UI, app, poster, layout, typography, mockup

Step 2: Load Genre-Specific Patterns

Read ${CLAUDE_PLUGIN_ROOT}/data/prompts-by-category.json and extract techniques from top prompts in that genre.

Top techniques by genre:

• Food: overhead view, macro closeup, steam visualization, warm lighting
• Portrait: depth of field, bokeh, skin texture detail, artificial lighting
• Product: centered composition, minimalist style, studio lighting, grid layouts
• 3D: array/list format, negative constraints, white background, icon grids
• Cinematic: volumetric lighting, hard contrast, dramatic composition
• Design: minimalist style, geometric shapes, limited color palette

Step 3: Apply the 6 Rules with Genre Context

1. Identify genre from user's request
2. Load top techniques from that genre's viral prompts
3. Extract the core subject they want to depict
4. Apply Rule 1: Replace feeling words with genre-appropriate pro terms
5. Apply Rule 2: Add genre-typical technical parameters
6. Apply Rule 4: Layer in genre-appropriate sensory descriptions
7. Apply Rule 3: Add genre-specific negative constraints
8. Apply Rule 5/6: Structure using genre's preferred format

Quick Optimization Process (Simplified)

1. Identify scene type from the user's request
2. Extract the core subject they want to depict
3. Apply Rule 1: Replace feeling words with pro terms
4. Apply Rule 2: Add quantified technical parameters
5. Apply Rule 4: Layer in sensory descriptions
6. Apply Rule 3: Add negative constraints
7. Apply Rule 5/6: Structure appropriately for complexity

Example Transformation

User input: "a bowl of ramen"

Optimized output:

Steaming bowl of authentic Japanese tonkotsu ramen, rich milky pork bone broth.

Visual Style:
High-end culinary magazine aesthetic. Warm earth tones with cream and amber highlights. Shot with Hasselblad medium format quality.

Composition:
45-degree overhead angle, 85mm lens, f/2.8, shallow depth of field on the soft-boiled egg.

Sensory Details:
Steam wisps rising and curling, noodles glisten with broth, chashu pork with caramelized edges, the aroma of garlic and sesame seems to penetrate the frame.

Constraints:
No utensils in frame. No text or watermarks. Maintain appetizing warm color temperature.

Additional Resources

Reference Files

For detailed patterns and terminology, consult:

• references/scene-guide.md - Scene-specific optimization patterns
• references/terminology.md - Complete professional terminology banks
• references/top-prompts.md - Top 50 viral prompts for inspiration

Generation Script

Execute image generation:

node "${CLAUDE_PLUGIN_ROOT}/scripts/gen.js" "YOUR_OPTIMIZED_PROMPT"

Model Information

This skill is optimized for Gemini 3 Pro Image (gemini-3-pro-image) via the antigravity-claude-proxy at localhost:8080.

The model excels at:

• Photorealistic imagery
• Complex scene composition
• Following detailed structured prompts
• Respecting negative constraints