Hello,
you might know some of the tutorials I wrote for collaborating with ChatGPT on songs and tracks. (If not, you can check them here: https://laibyrinth.blogspot.com/p/how-to-create-music-with-chatgpt.html )
While the response to the tutorials has been overwhelming (to say the least), *direct* feedback to the tutorials has been sparse;
People told me the tutorials were inspiring and of great use. But did anyone actually produce a full track this way?
I got a lot of feedback in the way of "I wrote 90% of a track according to your tutorial, but then ChatGPT failed at [task x]". I tried to suggest solutions to that specific error, but usually got no further feedback whether it actually worked or not.
And more stuff like that...
But to cut things short: I want to further improve the tutorials, and the connected ChatGPT music productions technique, but I think I need the feedback and help by others to do that in a meaningful way!
As you can get only so far on your own...
Before I go on with my questions, I'll give you a short TL;DR, in case you didn't read the tutorials, or to bring them back to memory.
The main idea is that you do not use ChatGPT for sound synthesis itself, like most other "AI projects" do, but instead connect ChatGPT to your DAW.
ChatGPT tells you all the notes, patterns, sequences, mixing and production notes, you put these to your DAW.
ChatGPT is the producer and writes the track - you are the mixing engineer and work the DAW for ChatGPT.
The original idea was to get a finished track as a result - and yes, this really works, and I already produced a few tracks this way.
But there are also tutorials in this series that are about generating specific elements of a track only.
So here are some thoughts on issues and aspects of the tutorials that I already came up with:
1. Is it more useful to find ways to produce a complete track - or to work with ChatGPT on very specific element only? The latter case could, for example, mean that you already worked on a Techno track, but still lack a bass-line, good percussion sequences, effect, and so on... and ChatGPT fills in there.
2. Is it more useful if ChatGPT writes the track 100% start to finish - or more like an "executive producer" in the background that chimes in some general ideas, and corrections, now and then?
3. Some complained that using ChatGPT to produce music in genres like Techno or Acid Trance (as I did) would be "bullshit" - because these genres already exist - and it would be more interesting to see an AI produce music in a style that was unthinkable before the advent of AI.
What are your thoughts on that?
4. Automation. I deliberately designed this technique against automation. Which means that you have to type in all the notes, rhythms, etc. that ChatGPT gives you "by hand", turn all the (virtual) knobs with your (mouse) fingers.
I come from the 90s when producing music still meant connecting a lot of wires,´and equipment, or sitting 90 minutes next to your tape deck when recording a demo (and making sure you reversed the tape after 45 minutes).
So I see no problem in typing "numbers and letters" that ChatGPT gives you into your DAW for on an on (and on).
But I reckon that other people might not like it.
And let's face it: if you could "export" the notes and things gives you into ChatGPT by a mere click of
a mouse button, that would be a huge "game changer" and might get this AI music production technique extremely popular and big.
But the question then would be: is it better be "popular huge", or to stay "underground"?
5. The tutorials focus mostly on the creation of everything related to notes and patterns, such as melodies, harmonies, percussion sequences, and so forth.
But maybe people could also need more advise on other things - like mix-down, EQ filter settings, mastering, etc.
6. Connected to this is: should ChatGPT focus on these very direct, and concrete things, or not on stuff that is more abstract, conceptual?
For example: "Tutorial for creating a track based on a poem" or "Tutorial for creating a somber ambiance in your track regardless of the actual genre".
7. When using the tutorials, ChatGPT always added that, while it is happy to generate notes, drum sequences, etc. for me/us/everyone, the "human" producer sitting in front of the screen should not forget to add his own creativity, and add his own ideas to ChatGPT's output as well.
I mostly ignored this advise, and still think the tracks sound great. ChatGPT *is* capable of writing wonderful tracks start to finish, without human "correction".
Still, might it be more useful to address this in a tutorial, too? To point out how producers could build on ChatGPT's ideas?
Okay, I think these are enough questions for now.
So, if you want to help me with these tutorials, and this "special" way to write music together with AI, I'd be very grateful if you would give me some feedback related to one or two of these questions.
Or general feedback - did you like the tutorials? Did you manage to write a track this way? What worked well, what didn't?
And, most importantly: what are your own thoughts? your own suggestions?
What do you think could still be improved?
No comments:
Post a Comment