| name | media-prompt-craft |
| description | Craft effective prompts for AI image/video generation and search queries for stock photo services. Translates design direction and styleguide choices into specific, style-consistent prompts. Use when the user asks to "find images for my site", "generate a hero image", "what kind of images should I use", "create image prompts for my project", "write prompts for image generation", "Unsplash query for...", or when building a media asset list for a design project. Works with any styleguide and delegates actual generation to media-plugin.
|
Media Prompt Craft
The bridge between design intent and media execution. Translates styleguide choices into specific, style-consistent prompts for image/video generation and stock photo search queries.
Operates inside the brief. This skill is the prompt/query craft used at steps 5–6 of the visual-planning gate (for image, video, and stock-photo assets). Run visual-planning first to clarify the ask, lock the style, and decide what's IN/OUT — then use the techniques here to turn that into a precise prompt.
When to Use
- User needs images for a website, app, or presentation
- User has a styleguide and needs matching media assets
- User asks "what images should I use for my landing page?"
- User wants to generate hero images, illustrations, or backgrounds
- User needs Unsplash/stock photo search queries
- User wants consistent visual style across multiple images
When NOT to Use
- User wants to generate the actual image → use media-plugin/image-generation
- User wants to search stock photos directly → use media-plugin/image-sourcing
- User hasn't defined a design direction yet → use styleguide skill first
- User wants to fetch SVG icons → use media-plugin/icon-library
Prompt-from-Styleguide Workflow
Before writing prompts, extract the visual DNA from the project's styleguide:
- Color keywords — "warm earth tones", "cool blues and slate", "neon on dark"
- Mood — "serene and calm", "energetic and bold", "professional and trustworthy"
- Photography style — "editorial", "candid", "moody", "bright and airy"
- Composition preferences — "minimal negative space", "dynamic angles", "centered subjects"
- Subject matter — "people", "abstract", "landscapes", "technology"
These become your style prefix — a reusable string that ensures visual consistency.
Prompt Anatomy
Every effective image prompt follows this structure:
[Subject] + [Composition] + [Lighting] + [Color palette] + [Style] + [Mood] + [Technical] + [Negative]
| Component | What it controls | Example |
|---|
| Subject | What's in the image | "a woman working at a standing desk" |
| Composition | How it's framed | "shot from slightly above, rule of thirds" |
| Lighting | Light quality and direction | "soft natural window light, golden hour warmth" |
| Color palette | Color characteristics | "warm earth tones, muted terracotta and sage" |
| Style | Visual treatment | "editorial photography, shallow depth of field" |
| Mood | Emotional quality | "calm, focused, intentional" |
| Technical | Resolution and format | "16:9 aspect ratio, high resolution" |
| Negative | What to avoid | "no stock photo feel, no forced smiles" |
Style Consistency System
Define a style prefix that encodes the project's visual DNA, then append subject-specific details for each image:
STYLE PREFIX (reuse for every image):
"Editorial photography, warm earth tones, natural golden hour lighting,
shallow depth of field, grain texture, muted color grading"
IMAGE 1 — Hero:
[style prefix] + "wide shot of a modern coworking space, plants and
natural materials, people collaborating casually, 16:9"
IMAGE 2 — Feature card:
[style prefix] + "close-up of hands typing on a laptop, coffee nearby,
wooden desk surface, 3:4"
IMAGE 3 — Team section:
[style prefix] + "candid portrait of a smiling professional, neutral
background, waist-up framing, 1:1"
The style prefix ensures all images feel like they belong together.
Stock Photo Query Strategy
Stock photo searches require different techniques than AI generation prompts.
Query Construction Formula
[Subject] + [Mood/Setting] + [Style modifier]
Bad queries (too generic):
- "office" → returns generic stock
- "team meeting" → forced smiles, staged poses
- "technology" → blue glowing screens
Good queries (specific and styled):
- "candid coworking warm natural light" → authentic workspace
- "team collaboration casual modern office plants" → natural team shots
- "developer laptop minimal desk setup" → genuine tech imagery
Keyword Layering
Layer keywords to narrow results:
| Layer | Purpose | Examples |
|---|
| Subject | What to find | workspace, portrait, cityscape |
| Mood | Emotional quality | warm, dramatic, serene, energetic |
| Setting | Environment | outdoor, studio, minimal, urban |
| Style | Visual treatment | aerial, close-up, silhouette, flat-lay |
| Color | Palette hint | dark, pastel, monochrome, earth-tones |
Example: "aerial cityscape dusk warm tones" > "city at night"
Unsplash-Specific Tips
- Use adjectives more than nouns — "serene mountain lake" not just "lake"
- Combine mood + subject — "cozy reading corner" not "bookshelf"
- Add material/texture keywords — "concrete texture", "marble surface"
- Specify "no people" or "with people" for better results
- Use size parameters:
?w=1440&q=80 for hero images, ?w=640&q=80 for cards
See references/search-strategy.md for advanced query patterns.
Media Type Patterns
Hero Backgrounds
For AI generation:
"Abstract [mood] background, [color palette keywords], flowing organic shapes,
subtle gradient transitions, no text, wide format 16:9, high resolution"
For stock photos:
Query: "[mood] abstract background [color] minimal"
Size: ?w=1920&q=85
Feature Illustrations
For AI generation:
"[Style] illustration of [concept], [color palette], clean composition,
[flat/3D/isometric] style, white or transparent background, centered, 1:1"
For stock photos:
Query: "[concept] icon illustration [style] minimal"
Size: ?w=800&q=80
Team/People Photos
For stock photos (preferred over AI for people):
Query: "candid professional portrait [setting] [mood] natural light"
"diverse team collaboration [setting] authentic"
Size: ?w=400&q=80 (avatars), ?w=800&q=80 (team section)
Product Mockups
For AI generation:
"Clean product mockup, [device type] displaying [UI description],
[angle — isometric/front/perspective], [background — gradient/desk/minimal],
soft shadow, photorealistic, high resolution"
Abstract Textures & Patterns
For AI generation:
"Seamless [texture type] pattern, [color palette], subtle,
tileable, [material — paper/fabric/marble/concrete], no text"
For stock photos:
Query: "[material] texture [color] background minimal"
Size: ?w=1920&q=85
Video Backgrounds
For AI generation (via media-plugin/video-generation):
"Slow-motion [subject], [mood], [color grading], cinematic,
subtle movement, seamless loop, no text, 16:9, 10 seconds"
Style Vocabulary Quick Reference
See references/style-vocabulary.md for the comprehensive vocabulary. Key terms:
Lighting: golden hour, rim light, soft diffused, dramatic high-contrast, backlit, overcast flat
Composition: rule of thirds, centered, negative space, leading lines, frame within frame, aerial/bird's eye
Mood: serene, dramatic, whimsical, austere, intimate, epic, cozy, energetic
Color: muted, saturated, pastel, monochromatic, complementary, analogous, earth-tones
Style: photorealistic, flat illustration, isometric, watercolor, 3D render, editorial, cinematic
Cross-Plugin References
- media-plugin/image-generation — use
mcp__media-mcp__generate_image with the crafted prompts
- media-plugin/image-sourcing — use with the crafted search queries for Unsplash/Pexels/Pixabay
- media-plugin/video-generation — use
mcp__media-mcp__generate_video for video backgrounds
- media-plugin/icon-library — for SVG icons (don't generate these, fetch pre-made ones)
- styleguide skill — create the design language FIRST, then use this skill to translate it into prompts
Tips
- Always craft a style prefix first — consistency matters more than individual quality
- Stock photos for people, AI generation for abstract/conceptual — AI struggles with natural-looking humans
- Include negative prompts to avoid common AI artifacts: "no watermark, no text, no distortion"
- Test your Unsplash queries before committing — search results vary significantly with keyword order
- For presentations, reference office-plugin/pptx for recommended image dimensions per slide type
