*

The Main branch relies on the paid OpenAI API to work. This has been resolved (not released yet❗) by implementing LlamaCpp for locally hosted Langchain agents instead. For setup refer to the main repo and installation guide

• access YouTube as provider of video assets, surpassing stock footage by large
• avoids bans by using download streams with PyTube
• avoids YouTube v3 API's rate limit and auth
• avoids Pexels hard rate limit

Inside api_utils functions provide utility for working with different APIs. Files: image_api.py, pexels_api.py, youtube_api.py and eleven_api.py. We added youtube_api.py to source video assets directly from YouTube. Feel free to modify the _generateVideoUrls function for the hierachy of video asset sources.

• search_videos_YouTube(query_string)

  The search_videos_YouTube function takes a query string as input and searches for a video on YouTube based on that query. It returns the URL of the first search result if found, or None if no video is found.

  Integration of shortGPT.api_utils.youtube_api, which serves as an additional source for gathering footage based on the queried keyword. This is particularly useful when dealing with niche / comedy / meme topics where stock footage is not available. If nothing matches we backtrack to the pexels API.

• triage_videos_Youtube(expected_score_parameter) ❗not released yet

demo_new shows the accuracy of the newly added youtube query policy without further guidance, backtesting or content analysis. This can be improved by adding a content triage based on Clip2 and transcript analysis.

deprecated_demo shows the accuracy of Google's YouTube v3, you can find it here: https://vimeo.com/851101834?share=copy.