SuperAPI AI Directory

slide-deck-images

Design & creation

View repository

Updated 10-15 16:46

About this skill

Source: https://github.com/aipoch/medical-research-skills

When to Use

Use this skill when you need to produce slide images as SVG files (plus outline/prompts), especially when:

  1. A user asks for a “slide deck”, “PPT”, or “presentation” but explicitly wants images (SVG) rather than PPTX/PDF.
  2. You need a readable, shareable deck optimized for scrolling/swiping (each slide stands alone without a presenter).
  3. You want a style-driven deck using presets (e.g., corporate, minimal, blueprint) or custom style dimensions.
  4. You need a structured workflow with optional review gates (outline review, prompt review) before generating images.
  5. You need to regenerate specific slides (e.g., only slide 3 or slides 2,5,8) without rebuilding everything.

Key Features

  • SVG-only output: Generates .svg slide images (no PPTX/PDF generation).
  • Deterministic file structure:
    • Output directory: slides-svg/
    • Naming: NN-slide-[slug].svg
  • End-to-end workflow: Analyze → confirm → outline → prompts → SVG render → summary.
  • Style system:
    • Preset styles (e.g., blueprint, corporate, sketch-notes, notion, etc.)
    • Custom style dimensions: texture, mood, typography, density
    • Automatic preset recommendation based on content signals
  • Review controls:
    • --outline-only, --prompts-only, --images-only
    • Optional outline/prompt review steps via confirmation questions
  • Selective regeneration: --regenerate <N> or --regenerate 2,5,8
  • No image-model dependency: SVG rendering is performed by a Python script using configuration parameters.

Dependencies

  • Python 3.10+
  • Project script: scripts/generate_svg.py (invoked locally)
  • Reference/spec files (used by the workflow):
    • references/analysis-framework.md
    • references/outline-template.md
    • references/base-prompt.md
    • references/layouts.md
    • references/design-guidelines.md
    • references/dimensions/*.md
    • references/styles/<style>.md
    • references/config/preferences-schema.md

Example Usage

CLI (skill invocation)

/slide-deck "Explain OAuth 2.0 for product managers"
/slide-deck "Explain OAuth 2.0 for product managers" --style corporate
/slide-deck "Explain OAuth 2.0 for product managers" --audience executives
/slide-deck "Explain OAuth 2.0 for product managers" --lang en
/slide-deck "Explain OAuth 2.0 for product managers" --slides 10
/slide-deck "Explain OAuth 2.0 for product managers" --outline-only
/slide-deck "Explain OAuth 2.0 for product managers" --prompts-only
/slide-deck "Explain OAuth 2.0 for product managers" --images-only
/slide-deck "Explain OAuth 2.0 for product managers" --regenerate 3
/slide-deck "Explain OAuth 2.0 for product managers" --regenerate 2,5,8
/slide-deck  # then paste content

Generate SVG files locally

The SVG renderer reads parameters from its configuration area and does not use CLI arguments:

python scripts/generate_svg.py

Expected outputs:

  • outline.md
  • prompts/ (one prompt file per slide)
  • slides-svg/NN-slide-[slug].svg

Implementation Details

Output constraints

  • Format: .svg
  • Output directory: slides-svg/
  • Filename pattern: NN-slide-[slug].svg

Style system

Presets

Preset Characteristics Typical use
blueprint (default) grid + cool + technical + balanced architecture, system design
chalkboard organic + warm + handwritten + balanced education, tutorials
corporate clean + professional + geometric + balanced pitches, proposals
minimal clean + neutral + geometric + minimal executive briefings
sketch-notes organic + warm + handwritten + balanced education, tutorials
watercolor organic + warm + humanist + minimal lifestyle, wellness
dark-atmospheric clean + dark + editorial + balanced entertainment, gaming
notion clean + neutral + geometric + dense product demos, SaaS
bold-editorial clean + vibrant + editorial + balanced launches, keynotes
editorial-infographic clean + cool + editorial + dense explainers, research
fantasy-animation organic + vibrant + handwritten + minimal storytelling
intuition-machine clean + cool + technical + dense academia, technical docs
pixel-art pixel + vibrant + technical + balanced gaming, dev talks
scientific clean + cool + technical + dense medical, biology, chemistry
vector-illustration clean + vibrant + humanist + balanced creative, kids content
vintage paper + warm + editorial + balanced history, culture

Style dimensions

Dimension Options Meaning
Texture clean, grid, organic, pixel, paper background/texture treatment
Mood professional, warm, cool, vibrant, dark, neutral tone and color temperature
Typography geometric, humanist, handwritten, editorial, technical heading/body typographic feel
Density minimal, balanced, dense information density per slide

Full dimension specs: references/dimensions/*.md

Automatic preset recommendation (signals → preset)

Content signals Preset
tutorial, learn, education, guide, beginner sketch-notes
classroom, teaching, school, chalkboard chalkboard
architecture, system, data, analysis, technical blueprint
creative, children, kids, cute vector-illustration
briefing, academic, research, bilingual intuition-machine
executive, minimal, clean, simple minimal
saas, product, dashboard, metrics notion
investor, quarterly, business, corporate corporate
launch, marketing, keynote, magazine bold-editorial
entertainment, music, gaming, atmospheric dark-atmospheric
explainer, journalism, science communication editorial-infographic
story, fantasy, animation, magical fantasy-animation
gaming, retro, pixel, developer pixel-art
biology, chemistry, medical, scientific scientific
history, heritage, vintage, expedition vintage
lifestyle, wellness, travel, artistic watercolor
Default blueprint

Design philosophy (reader-first decks)

Decks are designed for reading and sharing, not live presenting:

  • Each slide is understandable without narration.
  • Logical flow when scrolling/swiping.
  • Each slide includes necessary context.
  • Optimized for social sharing.

See: references/design-guidelines.md (visual hierarchy, density, colors, fonts, recommendations).
Layouts: references/layouts.md.

Workflow and state files

Step 1: Setup & Analyze

  1. Load preferences from EXTEND.md (if present) and summarize:

    • Style (preset/custom)
    • Audience (or auto-detect)
    • Language (or auto-detect)
    • Review enabled/disabled
      Schema: references/config/preferences-schema.md
  2. Analyze content (per references/analysis-framework.md):

    • Save input as source.md (backup existing as source-backup-YYYYMMDD-HHMMSS.md)
    • Detect language
    • Recommend style preset from content signals
    • Estimate slide count
    • Generate topic slug
  3. Check existing content (required) before confirmation:

    • If output directory exists, ask how to proceed:
      • Regenerate outline (keep images)
      • Regenerate images (keep outline)
      • Backup and regenerate all ({slug}-backup-{timestamp})
      • Exit
    • Save analysis.md with topic, audience, signals, recommended style, slide count, language.

Step 2: Confirmation (required)

  • Two rounds:

    • Round 1: always
    • Round 2: only if “Custom dimensions” is selected
  • Round 1 questions (AskUserQuestion):

    1. Style (recommended preset / alternative / custom dimensions)
    2. Audience (general/beginners/experts/executives)
    3. Slide count (recommended / fewer / more)
    4. Review outline? (yes/no)
    5. Review prompts? (yes/no)
  • Round 2 (custom dimensions): AskUserQuestion for texture, mood, typography, density.

After confirmation:

  • Write results back to analysis.md
  • Record skip_outline_review and skip_prompt_review

Step 3: Generate outline

  • If preset: read references/styles/{preset}.md
  • If custom: combine from references/dimensions/
  • Follow references/outline-template.md
  • Save as outline.md
  • Stop if --outline-only
  • If skip_outline_review → skip to Step 5, else Step 4

Step 4: Review outline (conditional)

  • Show slide summary table (title/type/layout)
  • Ask whether to proceed, edit outline.md, or regenerate outline

Step 5: Generate prompts

  • Base prompt: references/base-prompt.md
  • For each slide:
    • Extract style instructions from outline
    • Merge layout guidance from references/layouts.md when specified
  • Save to prompts/
    • Backup existing prompt files as prompts/NN-slide-{slug}-backup-YYYYMMDD-HHMMSS.md
  • Stop if --prompts-only
  • If skip_prompt_review → skip to Step 7, else Step 6

Step 6: Review prompts (conditional)

  • Show prompt list table and prompts/ path
  • Ask whether to proceed, edit prompts, or regenerate prompts

Step 7: Generate SVG images

  • If --images-only, start here
  • If --regenerate, render only specified pages
  • Render via scripts/generate_svg.py
  • Create session id: slides-{topic-slug}-{timestamp}
  • Backup existing images as NN-slide-{slug}-backup-YYYYMMDD-HHMMSS.svg
  • Progress reporting: Generated X/N (in user language)
  • Retry once on failure, then report error

Step 8: Output summary

Include:

  • Topic
  • Style (preset or custom dimensions)
  • Total slides
  • List of generated SVG filenames
  • outline.md location