slidebake 0.1.1

Bake slide-deck PDFs into page-by-page Markdown with OCR and optional translation.

slidebake

slidebake turns slide-deck PDFs, including image-only PDFs exported from PowerPoint, into page-by-page Markdown. It uses local OCR first, then can ask the OpenAI API to clean OCR noise and translate each page.

Install

slidebake should be installed with Python 3.12. One of its OCR dependencies does not currently support newer Python versions.

pip install slidebake

For local development:

uv venv --python 3.12
uv sync
uv run slidebake --help

Usage

slidebake slides.pdf
# writes slides.md

slidebake slides.pdf --target-lang zh-CN
# writes slides_zh-CN_translated.md

slidebake slides.pdf --target-lang zh-CN --bilingual
# writes slides_zh-CN_bilingual.md

Use -o/--output to choose an exact output path instead.

Translation requires:

export OPENAI_API_KEY="your_api_key_here"

You can also put OpenAI or OpenAI-compatible provider settings in ~/.config/slidebake/config.toml:

[openai]
api_key = "your_api_key_here"
model = "gpt-5.4-mini"
api = "responses"

For providers that expose an OpenAI-compatible Chat Completions endpoint:

[openai]
api_key = "provider_api_key"
base_url = "https://api.example.com/v1"
model = "provider-model"
api = "chat_completions"

Command-line options and SLIDEBAKE_OPENAI_* environment variables override the config file. Use --config path/to/config.toml to load a different config file.

Useful options:

• --pages 1-10,15 limits processing to specific pages.
• --dpi 220 changes render resolution for OCR.
• --model gpt-5.4-mini changes the OpenAI model.
• --openai-base-url https://api.example.com/v1 uses an OpenAI-compatible endpoint.
• --openai-api chat_completions uses the Chat Completions compatibility path.
• --check-key checks resolved OpenAI-compatible settings without processing a PDF.
• --keep-temp keeps rendered page images for debugging.
• --overwrite replaces an existing output file.