YouTube video input for Gemini models
Gemini 2 flash and 2.5 pro bought have support YouTube video input over the API.
Sometimes relevant information can only be found in video.
The ability to accurately and quickly summarize videos.
Being able to show when and where charts and graphics are displayed.
Log in to comment and vote
Comments3
Lime Sunbeam
Merged request
Automatic YouTube Video Transcripts
When pasting a video link, automatically get the transcript for the video to use as context for the prompt. I often use T3 Chat for summaries and key points of YouTube videos that I don’t want to watch or sift through. I currently rely on a third party tool for it which breaks flow. This would be a game changing feature for me and make it much more seamless to just directly paste YouTube or other video links into T3 Chat.
Salmon Jaguar
Reasoning about a YouTube video is the core feature here for me (any model is fine):
Use-cases:
Determine if longer videos or content from unknown-to-me authors is worth the watch.
Ugly, badly edited video may have the most worthwhile and meaningful content. Researchers and the like are notoriously bad at making content - look at all the “what is CSS” unstyled websites, with truly brilliant content. I want to know via AI analysis and/or summerization if it is: brilliant (for me), or just bad in style and content.
Clickbait disarming (sorry Theo). Worth the watch Y/N… can be difficult to tell, even for known authors.
The Gemini models may just be the most practical ones due to YT ToS, otherwise you would need to negotiate a contract with G**gle to be allowed to scrape & use content…
As an aside, it might be interesting to investigate how
https://www.getrecall.ai/ and similar tooling does this, both technically and probably more important legally (YT ToS).
Azure Leaf
This would be such a killer feature. Hate that I have to rely on gemini.google.com right now for being able to summarize YouTube Videos…