PDF to Anki

All credit goes to benno094, who originally created the code.

Introduction

PDF to Anki program using GPT 4o mini from OpenAI. Streamlit is the web-GUI. Shout-out to OpenAI and Streamlit for saving me a ton of work!

Requirements:

• An OpenAI API key
• Anki running with AnkiConnect installed (Addon #: 2055492159)
• Not needed, but highly recommend AnKing Notetype (Addon #: 952691989)

If compiling to run locally:

• An OpenAI key
  • Needs to be added as OPENAI_API_KEY="[Key here]" in .streamlit\secrets.toml

Usage:

1. Add an OpenAI API key.
2. Choose a PDF file.
3. Select a page range and a deck for the flashcards to be added to.
4. Flash cards are generated for each page and can then be modified before being added to Anki.

Note: App adds a custom note type (AnKingOverhual) so that there is no issue with note's name and fields being in another language. I still recommend you install AnKing Notetype (Addon #: 952691989) for more control

To do:

• Add option to log in to save preferences
• Check if dependencies are all really needed
• Allow user to change prompt options
• Error handling of api errors

Changelog:

0.83 Beta

• Images are now properly imported and stored in the corresponding Anki Media Collection folder
• Images are no longer Base64, and are instead proper jpg files
• Now compatible with Ankihub (Ankihub is incompatible with Base64 images)
• Used FirstAid and Anking as examples to improve the prompt. Will add more later
• Decreased size of images, from full sized images to truncated DPI = 150
• Fixed some bugs with card importing and added more debugging errors
• Images aren't analyzed by gpt 4o mini anymore. Too many tokens and not worth the effort for every page
• Notetype is now AnKingOverhaul

0.82 Beta

• Made it compatible with Cloze deletions
• Prompt automatically produces clozes (1-2 average per card)
• Fixed the Notetype, and made it compatible with AnKing's Notetype
• Fixed the naming convention of the cards

0.81 Beta

• Changed gpt 3.5 turbo for gpt 4o mini
• Images are now transferred with the cards
• Images are now analyzed by gpt 4o mini (not consistent)
• Prompt was improved, but orientated for medical students.

0.8 beta

• Add option to use GPT4
• Errors returned by openai are displayed
• Added button to hide page

0.7 beta

• Can now add one image per flashcard
• Can view text extracted from flashcard
• Added option to add new flashcard
• Modified handling of bad cards returned by GPT
• Flashcards now support Markdown and retain formatting provided by GPT
• Cards can now be readded

0.6 alpha

• Modified workflow to make more sense
• Unneeded elements now disappear
• App now adds new note type to Anki
• Language auto-detected
• Updated GPT prompt to use tool_call
• Added preview of file

0.51 alpha

• Decks can now be chosen

0.5 alpha

• Lowered chance that GPT returns non-parseable result by moving to functions
• Skip pages without relevant info

0.46 alpha

• Added JPEG compression to reduce image size
• Fixed key error causing Streamlit crash
• Stopped unnecessary regenerations of flashcards
• Automatically check for AnkiConnect connection
• Show decks

0.45 alpha

• Improved memory usage
• Can now add flashcards to Anki without running locally
• Added filename and page as tag
• Max. uploaded file size reduced to 50MB

0.4 alpha

• Can now add without running locally
• Can now add page range and change on-the-fly
• Option to select returned language

0.3 alpha

• User is updated on status and errors
• Improved GPT prompt

0.2 alpha

• Changed to Streamlit Web-GUI
• Removed image detection
• Shifted flashcards view to show alongside page preview
• Using pymupdf instead of pdf2image to reduce reliance on external libraries

0.1 alpha

• First release