slide-vision

Enrich markdown slide decks with images — from local folders, DALL-E 3, Pexels, or Unsplash.

Install

pip install slide-vision
# With DALL-E support:
pip install slide-vision[openai]

Quick start

# Create config
slide-vision init

# Enrich a slide deck (uses configured providers)
slide-vision enrich presentation.md

# Skip all APIs — just produce image prompts
slide-vision enrich presentation.md --source none

# Use local images + Pexels fallback
slide-vision enrich presentation.md --source local,pexels --local-dir ./images/

# Also produce a .zip
slide-vision enrich presentation.md --zip

Pre-process local images

Use scan to AI-rename your local images for better keyword matching:

slide-vision scan ./my-images/ --provider openai --dry-run
slide-vision scan ./my-images/ --provider openai

Output structure

presentation-enriched/
├── presentation.md    # enriched markdown with image tags
├── images/            # all sourced images
└── prompts.md         # AI prompts for slides without images (if any)

API keys

Check status: slide-vision providers

Custom providers

Bring your own image source or vision model — no fork required.

Quickest: a module:Attribute import spec. Point --source (or scan --provider) at any importable object. An image provider just needs a get_image(slide, output_dir) method; a vision provider is any callable(path) -> str.

# Your own image provider class
slide-vision enrich deck.md --source mypkg.providers:MyProvider

# Mix it with builtins — first hit wins
slide-vision enrich deck.md --source local,mypkg.providers:MyProvider --local-dir ./img

# Your own vision model for scan
slide-vision scan ./img/ --provider mypkg.vision:describe

Image provider contract (implements the ImageProvider protocol):

from slide_vision.providers.base import ImageResult

class MyProvider:
    name = "mine"
    def is_available(self) -> bool: ...
    def get_image(self, slide, output_dir) -> ImageResult | None: ...

Publishable plugins (entry points). Ship a package that registers under slide_vision.providers (images) or slide_vision.vision_providers (scan); once installed it's usable by name and shows up in slide-vision providers:

[project.entry-points."slide_vision.providers"]
gemini = "slide_vision_gemini:GeminiProvider"

[project.entry-points."slide_vision.vision_providers"]
claude = "slide_vision_claude:describe"

Programmatic registration (in your own scripts):

from slide_vision.providers.factory import register_provider
from slide_vision.scanner import register_vision_provider

register_provider("mine", lambda ctx: MyProvider())
register_vision_provider("mine", lambda path: "a caption")