Google Notebook LM
Not Real… by Jim Reed (Awarded Penny Johnson Trophy for Best Editing – 2024)
It’s a fun video, but contains deeper messages about Ai and reality.
I called it ‘Not Real…’ but it’s possibly easier to say what is real in the video:
- The opening words were written by me.
- Google really do have NotebookLM. It’s free to use, and can synthesize and analyse all manner of inputs – text, documents, papers, videos, audio. It has the added feature of generating an Ai two person ‘deep dive’ podcast from the inputs.
- I wrote a ‘press release’ (below) about a fictional company ‘Podcast Central Studios’ and added my comments to the presenters (also below) as my input into Google NotebookLM. I wanted to see how Ai would react to discussing ‘themselves’ as not being real, and wanted to provoke a discussion about Ai and reality.
I ran NotebookLM 10 times and downloaded the audio it produced. Each run took about 10-15 minutes and each produced slightly different results, ranging from 4 to 10 minute outputs – in some of the outputs the presenters referred to another studio having Ai hosts, not themselves, and expressed sympathy at them not realizing that their whole backstory was created by their studio.
The results were better than I had expected, and philosophically quite deep. Interestingly they also seemed to ridicule, or at least distance themselves from the notion that the podcast was Ai generated on the basis that the amount of work involved was very complex and would require large amounts of power.
All the outputs were interesting, but I picked the ‘chunks’ that dealt with them being Ai and the discussions about reality. The interactions and their repartee were genuine two handers – I bolted different ‘chunks’ from a selection of the 10 outputs together to make their section just over 4 minutes long.
I thought their discussions were very deep, very thoughtful, and very insightful. That they recognized the irony of the possibility that maybe they were just lines of code discussing their own existence was delightful!
The tone and inflection in the Ai Generated voices was impressive. Particularly when the presenter reads the words, he changes to use exactly the right tone for reading, which is very different from, for example, chatting. There was one really interesting clip I would have liked to have used (but it didn’t fit) when the female voice made an error and chose the wrong word. She spoke the word, there was an immediate ‘er’ as it recognized the wrong choice, and then spoke the correct word and carried on. It was so real. We have all been there – misspoken or mispronounced a word and then corrected it before continuing.
11Labs is one of the very best speech generation systems – I used it to clone my voice in my video ‘Move 37 : Dreaming of Electric Sheep’ when I didn’t actually say any words in that video – my voice was entirely generated from the text of my narrators script. But the Google generation is even more exceptional – just by listening it’s hard to know if the voices are genuine or not.
Finally, and for clarity, all the images were generated by Black Forest Labs Flux1.0; the music was generated by Suno; the opening voice was 11Labs; the Podcast hosts voices were Google NotebookLM, and the character ‘avatars’ (with lipsync) were from HeyGen.
Ai is a tool, in much the same way that the Lighting or Sound engineers use there equipment, and so I added the credit of Ai Engineer to the end of the video. Rather than stifle creativity, I think this video shows how Ai can enhance and encourage creativity. Ai is here, and it will only get better. As video makers we should embrace it, rather than shun it. It is the future of our video making.
The full text from the two documents that I uploaded to Google NotebookLM is below:
I am in awe at what can be achieved using Ai – and this is only the beginning – it will only get better!
Jim