RapidComposer v5 beta feedback and discussion

[lulukom]

[lovemusic15]

I agree with your point, Mike; you make a good observation.

Like you mentioned currently with AI art, You can describe precisely what you want, like an elephant flying on a plane diving into the sea, and that's pretty much what you'll get.

However, with RC currently, instructing it for a guitar bass line results in a random riff. Unlike other AI, I can't ask RC for a riff similar to Status Quo because it would require extensive training. Hence, I believe the best approach for RC and AI is to feed it what you want.

Software like RC and AI can quickly take you to a creative place, generating better ideas than you might have thought of yourself.

Looking at what people on YouTube and social media ask for from AI is crucial. As RC grows, it becomes more beneficial for everyone. Since anyone can access free MIDI files, there's no reason why we can't feed RC with what we like. I have thousands of MIDI files that I'm willing to share.

AI's great aspect is its ability to create original compositions every time. Therefore, focusing on AI with RC should involve more than just suggesting chord progressions. It should enable us to take our favorite piano piece, change the syncopation and BPM, play it in a style similar to Coldplay, and incorporate a chord progression from a Justin Timberlake song. ( random example ) The essence of music production lies in the imagination of the producer, spending less time on what's considered good and more time experimenting with what you already like.

Here's one suggestion on how this could work: just as a random example , take three different MIDI files from Status Quo, for instance. Choose three of your favorite chord progressions, specify the amount of rest and syncopation you'd like to be used. Have RC lock all these parameters from those songs, and then generate new compositions exclusively using those specified parameters. Think of it like blending your favorite colors into a paint tin and stirring it for a unique, harmonious result.

[Pjotter]

RC is far ahead of any other. But AI is growing fast so including AI in RC is perfect. Just found a list of small AI tools. Maybe it has ideas for RC.

https://www.audiocipher.com/post/ai-music-app

[Mykyndryd_]

I agree with your point, Mike; you make a good observation.

Thanks, but I would add a further caveat; that we have all an assumption that a good piece of music has another, "trojan horse" good piece within, which is uncovered by rearranging the original form or becoming inspired by such.

As I always say, the world's best Custard is not that great when poured on the world's best Salad

AI would have to be able to take the examples, compare it with other Quo stuff (of the same period etc) and then say "I actually get what's going on here", and in the case of Quo, bang a I IV V song out with two guitar parts as a blues shuffle

I feel we are inching towards where we need to be but it's asking the right question that will give the right answer and that is still a real mystery in some ways for music creation.

One of the problems is that we are looking for originality from something that has already been done which is the opposite way to how creativity works, composers tend not to look at the back catalogue to produce new stuff other than style influence. It's a real paradox in that sense.

Cheers. Mike.

[BluGenes]

Just so you guys know, we tested that approach already, months ago. Several problems lead us to decide against this approach. First, at the time, you needed two programming languages, Python and C++. The Python part would have required everyone to have that installed, really not a feasible idea. It was very very slow. It needs an insane amount of data to produce very minimal results. It needs millions of samples. Length of time to produce a "model" ran into days.

A model is what the AI uses to produce the requested information from.

In my tests, after 6 hours of training, it barely made any musical sense. After 12 hours, it did better, but, not useable. The files are HUGE too, in the giga bytes.

Unlike image AI, there are no outside available models to use that are open sourced. So, it would be up to you to create that.

Another really really REALLY big issue is copyright infringement.

[lovemusic15]

Hi Mike,

"I understand what you're saying to a certain extent, but you have to look at it another way. It's not about getting something entirely original from what has been done; it's about taking something you like as a starting point to create something new. It's a bit like taking all the cast from Harry Potter but getting Quentin Tarantino to direct his type of film; there may be similarities, but it's never going to be the same as Harry Potter the film. Another analogy is that we could all make Shepherd's pie using the same ingredients, but it doesn't mean it's going to taste exactly the same as other people's versions. Just like when someone does a cover song, think of someone like Drake singing a Spandau Ballet cover; he's going to give it some sort of originality or totally revamp the original idea. As I'm sure you're well aware, some of the biggest songs of all time were originally covers of an original song. The main point is, you're always going to get better results if you start off with something good."

[lovemusic15]

Hi BluGenes

"I completely understand your perspective regarding the training aspect. What I'm proposing is essentially an enhancement of what's already possible in RC. It requires an update and fixes to align with the approach I'm suggesting. Consider the Markov generator, for instance; you can input a MIDI file of your choice, rearrange it, and create new melodies. Unfortunately, due to the algorithm, it doesn't work well for custom phrases. On the other hand, MIDI Mutate excels in analyzing custom phrases.

Combining both functionalities could analyze the original shape and construct new ones based on those parameters. Crucially, it would ensure it doesn't recreate the exact pattern initially inputted, addressing concerns about copyright infringement since it would never replay the same melodies. In simpler terms, using the analogy of numbers like 12345, it could generate every different combination except those. Currently, you can only drop one MIDI file at a time, but with this enhancement, you'd be able to drop several and store them. This functionality would be akin to how a drum machine sampler works, allowing you to toss in all your favorite samples and create brand new drum kits."

[lulukom]

I see talk about some very complex ways of generating phrases.. But to be honest, I personally need one, well and variedly customizable machine to produce many variations of a midi file that I have played and recorded. If this is a slap bass phrase, then the program should give me, well, 100 phrases with all the shades that I put into the phrase and of course a lot of clear sliders to customize this. And I’ll especially add that it should be possible to pre-play these phrases in the entire set of scales corresponding to the chords of the master track. It's the same with chord progressions. I don't know if it's necessary to use AI like this? Despite all the enormous potential of the RC, I cannot make the work convenient, which I roughly described above..

[lulukom]

1. Crash when I click with the mouse on any slider in the head of the track. The report is attached.
And the second crash is not clear why.
2. Two notes on the Mutator.
- I did not find the opportunity to return to the original midi file, after various manipulations with it.
- Many algorithms create double or even triple notes, like in a chord..And if I make a monophonic phrase, then every time I have to go to the Chords tab and make it monophonic.

[BluGenes]

It would be nice if for the "additional input" prompt, have it keep anything new that you have typed. Or at least make that optional. After you have generated something, then re-open the dialog, it has the default text in the edit field again, which, if that is were you are working to refine a prompt, then it becomes tedious to change it..

also, my suggestion for the polyphonic stuff.. I was using the wrong GPT model, so, don't bother with that..

[lovemusic15]

I see talk about some very complex ways of generating phrases.. But to be honest, I personally need one, well and variedly customizable machine to produce many variations of a midi file that I have played and recorded. If this is a slap bass phrase, then the program should give me, well, 100 phrases with all the shades that I put into the phrase and of course a lot of clear sliders to customize this. And I’ll especially add that it should be possible to pre-play these phrases in the entire set of scales corresponding to the chords of the master track. It's the same with chord progressions. I don't know if it's necessary to use AI like this? Despite all the enormous potential of the RC, I cannot make the work convenient, which I roughly described above..

I believe we're on the same page because I'm not necessarily saying it needs AI. We just need the ability to adjust things the way we want. That's why it's essential to be able to input a phrase and have RC save those exact parameters, giving us all the options to change things as we want and create additional variations based on that. Like I keep saying, there are many features in RC; they just need to be put together in one device. However, nothing will work well unless the analysis of the notes is correct. Currently, it works like a randomizer with the generators we've got. We need the option to make our own syncopation. little modifiers and sliders on the settings to create the kind of results we are looking for. I've been using a lot of variations over the years, and this is how I figured out it could all work together a lot better than they do at the moment when used separately. That's why I definitely don't think AI is necessarily needed; it just needs a little bit of RC magic and to have it all created in one generator."

[lulukom]

+1

[lovemusic15]

"Here's a composite mockup integrating all the necessary elements to perfect the user experience for generating, recreating, and remixing custom phrases. It combines the phrase analysis algorithm from the MIDI mutator with the functionality of the Markov generator, enriched with additional features. The aim is to grant users complete control over the process."

[lovemusic15]

close up

[lovemusic15]