I would love to hear some suggested settings/hints/tips for the AI generator in Rapid Composer 5.
Can some people who are experienced with ambient music and Rapid Composer give me some AI pointers to auto generate instant creative stuff in the same vein as Hearts of Space Radio HOS.com or SpottedPeccary.com or Brian Eno with something like a Peaceful/Euphoric/Dreamy/Calming type feel or something to show me the kind of stuff Rapid's AI can do to introduce me to it's deep features?
I'm not sure how to make the AI compose a whole song with it's Type setting and Instructions entry. I know from the user guide I need to sign up on the ChatGPT page or TurboChatGPT and pay and then login and enter theory ideas but I am limited in theory ideas. Let's get this thread started people. The AI is a great feature!
I would love a few people to help me with the new AI feature for ambient.
So far all the generators used with the .sf2 format pianos and other general MIDI sounds are underwhelming for me in Rapid. Maybe someone can point me to other threads on RC's AI.
Rapid Composer 5 AI generator hints & tips thread
-
- KVRAF
- 1858 posts since 15 May, 2017
Right now, the AI generator works, but, in a very simplistic way. What you want is exactly what I am working on for it, but, has it's own design issues to be worked out. Also, Atilla might have this on the back burner for now too.
Hopefully I can have a video showing what is going on with this soon.
As an example, this is one of my "test prompts":
Generate a piano sonata in the style of Ludwig van Beethoven. This piece should not only reflect Beethoven's structural and thematic sophistication but also his ability to create rich, interlocking harmonies between multiple melodies.
I also found a serious pitfall with the AI, which is the cost. Granted, it's very cheap using the API, but, it could be very token heavy too, which would eat a lot of tokens between each session, eventually running out, very fast, of course.
As opposed to the web site subscription, were you have unlimited data, but, a very high subscription rate, but not the convenience of being built into an app, like RC.
Also, note too, I am not using RC to do this, but, I wrote my own app for this. However, it can be applied to RC, when Atilla can get back to work on this.
I kind of helped out by doing the research, so he didn't have too..
Hopefully I can have a video showing what is going on with this soon.
As an example, this is one of my "test prompts":
Generate a piano sonata in the style of Ludwig van Beethoven. This piece should not only reflect Beethoven's structural and thematic sophistication but also his ability to create rich, interlocking harmonies between multiple melodies.
I also found a serious pitfall with the AI, which is the cost. Granted, it's very cheap using the API, but, it could be very token heavy too, which would eat a lot of tokens between each session, eventually running out, very fast, of course.
As opposed to the web site subscription, were you have unlimited data, but, a very high subscription rate, but not the convenience of being built into an app, like RC.
Also, note too, I am not using RC to do this, but, I wrote my own app for this. However, it can be applied to RC, when Atilla can get back to work on this.
I kind of helped out by doing the research, so he didn't have too..
-
musicdevelopments musicdevelopments https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=223336 - KVRAF
- 5436 posts since 9 Jan, 2010
Thanks for this question, and thanks BluGenes, for the great reply.
You can get many kinds suggestions from AI, but getting good phrases are the trickiest. There is a big difference which model you use, e.g. GPT-3.5 is useful and very cheap for music theory-related questions, but almost unusable for phrase generation, for which you should select one of the GPT-4 models. I can imagine the soon-to-be-introduced GPT-5 model will be even more suitable for music.
As a general rule, you must include as many details with the prompts as possible. Simple, short prompts won't give good results (e.g. "create a bass line for heavy metal").
At the moment it is not possible to create multiple tracks with AI, but this is on my to-do list and will be implemented very soon. The syntax that RapidComposer uses when communicating with GPT models is very compact by design. This is important for not wasting tokens (that is your money).
If you find the included soundfonts underwhelming, you can use VST/VST3 instrument and effects.
Thanks,
Attila
You can get many kinds suggestions from AI, but getting good phrases are the trickiest. There is a big difference which model you use, e.g. GPT-3.5 is useful and very cheap for music theory-related questions, but almost unusable for phrase generation, for which you should select one of the GPT-4 models. I can imagine the soon-to-be-introduced GPT-5 model will be even more suitable for music.
As a general rule, you must include as many details with the prompts as possible. Simple, short prompts won't give good results (e.g. "create a bass line for heavy metal").
At the moment it is not possible to create multiple tracks with AI, but this is on my to-do list and will be implemented very soon. The syntax that RapidComposer uses when communicating with GPT models is very compact by design. This is important for not wasting tokens (that is your money).
If you find the included soundfonts underwhelming, you can use VST/VST3 instrument and effects.
Thanks,
Attila
