content-visuals

name	content-visuals
description	Generate image prompts and visuals for blog articles. Use when the user asks to "create images for an article", "generate blog visuals", "make a cover image", "create social media cards", "/content-visuals", or needs image prompts for cover heroes, Facebook cards, or Instagram cards matching article content.
version	0.2.0

Content Visuals

Generate brand-aware image prompts and visuals for blog articles. Supports 9 art styles, 3 moods, 3 image types, and pluggable generation backends.

Quick Start

Provide an article path and optional overrides:

/content-visuals path/to/article.md
/content-visuals path/to/article.md --style bold-graphic --mood energetic
/content-visuals path/to/article.md --preset data-vis
/content-visuals path/to/article.md --quick
/content-visuals path/to/article.md --types cover,instagram
/content-visuals path/to/article.md --model flash
/content-visuals path/to/article.md --model pro --types cover

Three Dimensions

Styles (9 available)

Style	Best For
flat-vector	Clean geometric — Policy & Compliance
isometric	2.5D technical — Infrastructure, supply chains
hand-drawn	Sketch-like — Practical guides, how-to
corporate-modern	Polished gradient — Most blog posts (default)
bold-graphic	High contrast — Urgent news, deadlines
watercolor	Soft organic — Success stories, community
minimal-line	Single-weight premium — Executive content
data-visual	Chart-inspired — Statistics-heavy articles
cultural-motif	Local patterns — Culturally-themed topics

Moods (3 available)

• professional — corporate, trustworthy, clean
• energetic — vibrant, dynamic, action-oriented
• warm — friendly, approachable, community-focused

Image Types (3 available)

Type	Dimensions	Aspect Ratio
cover-hero	1200 x 675 px	16:9
facebook-card	1200 x 628 px	1.91:1
instagram-card	1080 x 1080 px (or 1080 x 1350 px carousel)	1:1 or 4:5

Auto-Selection

When no style is specified, match based on content pillar:

Content Pillar	Style	Mood
Policy & Compliance	corporate-modern	professional
Funding & Financing	data-visual	professional
Industry Opportunities	isometric	energetic
Practical Guides	hand-drawn	warm

Presets

Shortcut combinations:

Preset	Style + Mood
standard	corporate-modern + professional
data-vis	data-visual + professional
friendly	hand-drawn + warm
impact	bold-graphic + energetic
cultural	cultural-motif + warm
clean	minimal-line + professional
tech	isometric + professional
creative	watercolor + warm
flat	flat-vector + professional

Compatibility Matrix

++ recommended, + works, - avoid:

Style	Professional	Energetic	Warm
flat-vector	++	+	+
isometric	++	++	-
hand-drawn	+	+	++
corporate-modern	++	+	-
bold-graphic	+	++	-
watercolor	-	-	++
minimal-line	++	-	+
data-visual	++	+	-
cultural-motif	+	-	++

Brand Integration

Load brand colors and typography from config/brands/_active.json → points to the active brand file in config/brands/. Use the brand's palette to tint generated visuals.

Workflow

1. Read article — parse frontmatter, extract key statistics, identify content pillar
2. Auto-select style — match pillar to style+mood (or use overrides)
3. Confirm with user (unless --quick) — present recommendation, wait for approval
4. Ask image count — how many of each type?
5. Generate prompts — assemble 4-layer prompt: type template + style rules + brand palette + article context
6. Generate images (if backend configured) — see sub-steps below
7. Save output — prompts and images to data/images/YYYYMMDD/{article-slug}/

Step 6: Image Generation Details

6a. Check Environment

Determine which backend is configured in config/image-generation.json.

If backend is prompt-only: Skip generation. Prompts are already saved from Step 5.

If backend is gemini:

1. Check if $GEMINI_API_KEY is set in the environment
2. If missing, tell the user:

   "Image generation requires a Gemini API key. Your prompts have been saved to data/images/... — you can copy them into Google AI Studio or any image generation tool to create your images. To enable automatic generation in Claude Code, set GEMINI_API_KEY in your .env file."

3. If present, proceed to 6b

If backend is any other value: Fall back to prompt-only and notify the user that the backend is not yet implemented.

6b. Select Model

Resolve the model ID from the --model flag:

• --model flash → gemini-2.5-flash-image (default if no flag)
• --model pro → gemini-3-pro-image-preview

Read model config from config/image-generation.json → backends.gemini.models.

6c. Generate Each Image

For each prompt file saved in Step 5, follow the procedure in references/backends/gemini.md:

1. Read the prompt markdown file (strip YAML frontmatter, use only the prompt body)
2. Build JSON payload with jq → temp file
3. Call the Gemini API via curl
4. Extract base64 image data with jq
5. Decode to PNG with base64 (platform-aware decode flag)
6. Validate file size with wc -c
7. Save to data/images/YYYYMMDD/{article-slug}/generated/{type}-{N}.png
8. Clean up temp files

6d. Handle Results

On success: Report each generated image with file path and size:

"Generated cover-hero.png (245,832 bytes)"

On failure (per image):

1. Log the error in the manifest
2. Confirm the prompt file is saved and available
3. Suggest the user can paste the prompt into Google AI Studio manually
4. Continue generating remaining images — do not abort the batch

On total failure (API key invalid, all requests fail):

1. Confirm all prompts are saved
2. Suggest manual generation path
3. Update manifest with error status

Output Structure

data/images/YYYYMMDD/{article-slug}/
  prompts/
    cover-hero-prompt.md
    facebook-card-1-prompt.md
    instagram-card-1-prompt.md
  generated/          (only when API backend is active)
    *.png
  manifest.json

Image Generation Backend

Load from config/image-generation.json. Available backends:

• prompt-only (default) — saves prompts only, no API calls
• gemini — Google Gemini (requires GEMINI_API_KEY)

If the selected backend fails or the API key is missing, fall back to prompt-only mode.

References

Detailed specifications for each dimension:

• references/styles/*.md — full style definitions for each of the 9 styles
• references/image-types/*.md — specs for cover-hero, facebook-card, instagram-card
• references/base-prompt.md — 4-layer prompt assembly guide
• references/auto-selection.md — content signal → style mapping
• references/compatibility.md — style × mood and style × image-type matrices
• references/style-presets.md — preset shortcut definitions
• references/visual-elements.md — universal visual rules
• references/workflow/*.md — step-by-step workflow procedures
• references/config/*.md — backend and preferences schemas
• references/backends/gemini.md — Gemini API curl implementation reference