By default, whenever I ask ChatGPT to write a post for social media, it's an extremely exaggerated post full of clickbait-y words and emojis. Saying "don't be cringe" helps with reducing that, though it's still not 100% perfect.
Saved the article. Thanks for putting this together.
I am familiar with almost all of the workflows and tools you are mentioning. I am trying much of what you are suggesting. It should be possible to get much better, not only more, done with it – in particular if the source material is good/rich.
Here are two things I can share:
1. Transcription
There is an app that locally transcribes audio with OpenAI's whisper model, called WhisperBoard. It runs on iOS, iPad and Mac. You can choose the model size and have some other settings. It is free and fantastic. Shout out to the developer!
my learning: you can't just drop your post in there. It is quite some work.
I have a Starter account. I played around with it a little, but a few days ago was the first time I used it to create an audio version of my last, very short post "Rocket Propulsion It Is."
So, I post in the article, and want to fix a missing paragraph or line break. I hit enter, and that unfortunately triggers the text-to-speech generations. You can't preview it. You can't stop it. Those credits were gone, by mistake. OK, no biggie.
Then the service, or the AI voice, was not able, despite various ways of me formatting it, to read the title of the post. It kept insisting that "Rocket Propulsion" was one thing, then it used the "It Is" from the title and blended that into the opening paragraph, which starts "Again I am delightfully surprised". Reading it as "It is again …"
So I had to re-generate the entire thing a few times. Lots of credits. No undo. I eventually had it read, or generate, the title and nothing but the title by itself, separately. There were a few other issues I had to fix in a similar fashion.
I also realized that the article, as it was written, did not work for reading. There were images in it. So I had to edit the article, adding descriptions of the images and some other adjustments.
The article burned up my entire monthly budget with Elevenlabs, even though it is a few minutes total length. I could not have generated any more fixes.
Thank you for sharing! I haven't seen WhisperBoard, I'll check that out.
I'd also agree with your notes on ElevenLabs - article and section titles are really difficult to do well! I think they're trying to address some of these issues with Projects (https://www.youtube.com/watch?v=DQWX21gmGEw), but unfortunately it's only available to paid users.
I'm very curious about the "don't be cringe" part of that first prompt. What is that doing?
By default, whenever I ask ChatGPT to write a post for social media, it's an extremely exaggerated post full of clickbait-y words and emojis. Saying "don't be cringe" helps with reducing that, though it's still not 100% perfect.
Fair enough! Simple instruction to try and tackle a real common problem.
Twelve Labs can also be a great tool for video summarization!
I haven't tried Twelve Labs much, I'll check it out.
Saved the article. Thanks for putting this together.
I am familiar with almost all of the workflows and tools you are mentioning. I am trying much of what you are suggesting. It should be possible to get much better, not only more, done with it – in particular if the source material is good/rich.
Here are two things I can share:
1. Transcription
There is an app that locally transcribes audio with OpenAI's whisper model, called WhisperBoard. It runs on iOS, iPad and Mac. You can choose the model size and have some other settings. It is free and fantastic. Shout out to the developer!
2. Elevenlabs.io
my learning: you can't just drop your post in there. It is quite some work.
I have a Starter account. I played around with it a little, but a few days ago was the first time I used it to create an audio version of my last, very short post "Rocket Propulsion It Is."
So, I post in the article, and want to fix a missing paragraph or line break. I hit enter, and that unfortunately triggers the text-to-speech generations. You can't preview it. You can't stop it. Those credits were gone, by mistake. OK, no biggie.
Then the service, or the AI voice, was not able, despite various ways of me formatting it, to read the title of the post. It kept insisting that "Rocket Propulsion" was one thing, then it used the "It Is" from the title and blended that into the opening paragraph, which starts "Again I am delightfully surprised". Reading it as "It is again …"
So I had to re-generate the entire thing a few times. Lots of credits. No undo. I eventually had it read, or generate, the title and nothing but the title by itself, separately. There were a few other issues I had to fix in a similar fashion.
I also realized that the article, as it was written, did not work for reading. There were images in it. So I had to edit the article, adding descriptions of the images and some other adjustments.
The article burned up my entire monthly budget with Elevenlabs, even though it is a few minutes total length. I could not have generated any more fixes.
Thank you for sharing! I haven't seen WhisperBoard, I'll check that out.
I'd also agree with your notes on ElevenLabs - article and section titles are really difficult to do well! I think they're trying to address some of these issues with Projects (https://www.youtube.com/watch?v=DQWX21gmGEw), but unfortunately it's only available to paid users.