Transcripts
3. Generating AI Music in V3: And here we are back on Suo I's website in case you
haven't been here before, it's Suoi, just type that in and it will
take you right there. And look here. Version three, Alpha access for
Pro and Premier. If you want to be a
free user on Suo do, of course you can
continue to do so. Version two is still awesome, but if you do have
a paid account, version three is now available. It's not a necessity, but I've already said what the benefits are and
if they suit you, then maybe paying for
it is now worth it. We're going to change
it right here, V two. We're going to click that and
change it to V three Alpha. It's changing every day. When they say they're changing every day, they are not lying. When I tried the Alpha the
day it came out, honestly, it was just the worst
songs are being made where they were 20
seconds long or there were lyrics half the time and then it was instrumental
half the time. Now only it's not been that many days and now
it's working a lot better. They are changing it
literally every day, updating it, But
let's give it a try. So we're going to put in just a general prompt
and see how they're doing in providing us what
they've claimed to provide us. Let's say a song about
learning on skillshare. So for this experiment, I'm not going to provide
any more information than that and see what
it comes up with, because that was
kind of my favorite thing about version two, was that you could put in a super simple
prompt like that. And I would get what? I just get something pretty awesome. Now one thing I'm noticing is that they op that
was pretty quick. One criticism I've
had so far is that they promised it would be
quicker than version two, but in my experience
it's been slower. But look, that was pretty quick. So very happy with that. Let's have a listen
put on your thinking. Cap fix time to expand to
learn what skills shake, you'll learn something new. One corset of time, Some create finding to graphic design to be ready
to unleash your potential. Leave the world behind. Level love en time to quo, you see you left. My impressions are the music is better and the vocal
sound the same to me. So the beat of the kick drum, the bass line,
they sound better. A big criticism of
my heart of version two was how bad the
kick drum sounded. If you took my previous class, you'll have seen that I
replaced the kick jump. I didn't replace it, there
was no way to do that. But I overlaid the
kick drum with my own kick drum sample to try to mitigate how
terrible the kick sounds so that the
song sounded so hollow with a really
fuzzy kick sound. It's no dancing to a fuzzy
kick sound. It can't be done. So I would overlay my own kick, but I think that sounds a lot better.
Let's try the other one. You got a dream deep down inside a pase can
no longer hide. You're searching
for a way to go and unlike the talent you
already know, thank you. Embrace the power. The world is to define cord to sing,
unlike your potential. Let your dreams take away Shale. I thought that was
awesome. I think honestly, skillshare could use that
as their commercial music. In my opinion,
obviously it's not as good quality yet as what a producer can
make on their own. But you know what, it
was pretty darn good. I think those male vocals
are sounding better. I thought the female vocals sounded the same as they
did in version two. I said in my last class, I didn't really like using
the male vocals at all. I thought that the male vocals, they just didn't sound as good. They sounded more
hollow and more fake. But listen to those male
vocals, they sounded great. So now maybe the male vocals are more viable than
the female ones, whereas in version two
female vocals were the star. Mm, how interesting. All right, now here's the real star of the update they
promised instrumental. How interesting. One comment I got on my previous
class on skillshare was, can they generate music and
then sing themselves over it? My answer was of course, you can take the vocals out with a plug in that will search and take out vocals
from a track. But God, does it
sound bad, Muddy, when you search for vocals
and take them out of a track, as low quality as what that was, I didn't recommend do it, but look at this, look at this improvement. Now, it's not such an issue. We can just make an
instrumental track. I'm going to ask for a bright
and cheery dance track. Now one thing Sono
have said is that you can put in BPM and
key signatures now. But that you have to expect
that it won't always work. It's, they're hopeful
that it will work. They've got things in place
to make it maybe work. Let's try a bright and cheery
dance track in C major at 120 BPM instrumental Create
if it can get this right. Now we knew in version two how we knew
that it wasn't working. Mainly was that the lyricist, the vocalist who's
singing will sing whatever the algorithm didn't
understand as instructions. I would put in major like this. Then the vocalist would
major in the lyrics and she would sing 120 BPM and then she'd rhyme
BPM with something. That's how we knew
it wasn't working. That was before even dragging it into Ableton and seeing
if it was right or wrong. Okay, so let's have a listen. With the vocals removed, it's a lot more obvious
that the music quality is still not something I
don't believe you could record yourself over that and
then put it on Spotify yet. It's definitely better
in my opinion here. If I go back to version two, play this Kyvat Pickle
like it's noble choice. No Cao's this sweet moment. Happiness is key,
so much see it. When it, on day
one I thought, oh, the improvement is
so little like God, how disappointing man do
I not feel that way now? Gosh, what a difference. See, that's quite
good. But anyway, I've got some more instructions
that are going to take advantage of some
of the new changes. So let's get a cup of coffee, and then, and then let's provide some more specific instructions in the continuation
of this track.
4. Working Around the Limitations to V3: One thing I've noticed
is with version two, I was able to click anywhere. In here, it was very
clear about how long the track was and I
could just click anywhere. And what I used
to like doing was clicking right near the end
to see how the ending was. But this one, it's saying
how far into the track I am, but it's not saying
how long it is. When I click, it
just doesn't work. That's interesting. I
wonder if this part of the algorithm still making the song as I play
it. I don't know. But they say that the
tracks are 2 minutes now. This is interesting because
this is going to change the way that we make
the track before. We would make them
30 seconds long. However many versions, six or seven or eight different
continuations to get to a full track length and then have the website
stitch them altogether. But if 2 minutes is quite long, we're going to have to use their new feature which allows you to continue
from any point. So I'm going to ask, I'm just
going to say a dance track because bright and cheery made those tracks a
little bit too kiddy. Dance in major. I'm going to write in here a two minute dance
track because that is the full length
apparently available. Now I have had some luck making two minute long tracks and I've had some situations where
they weren't quite 2 minutes, a two minute dance track
in C major at 120 Pm. All right, let's see
what we can get. Ordinarily in version two, you have to continue
from the 32nd track, was what used to be
the maximum length. At the end, you would
make a 32nd track, click continue, continue on, then you have a minute,
minute and a half, 2 minutes. Keep going until it's the
length that you want. And then click Get Hole Song, and it would stitch
it altogether. Now if we even do that once, the song is going to
be 4 minutes long, because it's 2 minutes
and 2 minutes, we're going to have to
use the new feature. I'm going to try to click again. Yeah, see no, can't do it. Really surprising. All right. I'm
going to write EDM. Really surprising that that
first one was what it was. I don't know who's
dancing to that. Goodness gracious also.
Isn't it funny how, I don't know if how much
you've played with this, but you're going to get
sunshine, electric pulse, neon nights, and something
about coffee always. Unless you change the title yourself. All right. Let's see. As you can hear, it's still
pretty bad at introductions. I thought that it
was quite nice. I thought that I
can't even rewind. What a shame. Quite good. The base drop is still
just so soft and hollow. I mean, look at what
I'm complaining about. The AI generation created something with a base
that's too hollow. Gosh, how spoiled can you be? You'd probably want to put
your own bass drop over top and the melodic lead line. God, could it benefit
from a bit of a reverb? But I can't put a
reverb on track. An individual stem within this AI can't be done,
which is a shame. But we're going to
continue this track and we're going to look
at the new features. Click that, See,
that's another thing. Within the first few days of
the Alpha, I noticed this. And it's still not fixed. It's still working. They promise faster generations. Now I think this is
how they're doing it in the old days
of version two. The old days a month ago, version two, they would generate
the whole track at once, and that's why we
could scrub around it. But it also meant it might
take a couple minutes. Now we're getting it quickly. Well we are now in
the first week. It was still taking
forever and it was still giving us this
pinwheel continue now, but it's just at the
expense of having it all done at once so it's
quicker but you know. Okay, what I'm going
to do is I'm going to click Continue
and I'm going to use the new feature
which is right here. Continue from, and
you need to write it in this exact format or
else it doesn't work. I'm going to put in 002, 39. So I'm going to continue it
from that base drop moment, which I remember being 39. I wish I could scrub back
and check, but I can't. Now, let's keep it going
as an instrumental. But what I am going to
do is I really want to see if I can add lyrics
to an instrumental track. So I'm going to do
that too. It's not working because I've
done something wrong. 39 That's right, can't be done. I wonder if instrumental
tracks can't be continued yet, turn off instrumental
still can't be continued. I wonder if it's
because it's not done. Let's try continuing from
our one from earlier, which I think this is the
one I liked, not that one. That's the one. And it might be just a case of
not being done. Can't be continued, is it? Because it's 2 minutes, maybe the 2 minutes
can't be continued. 0039 continue from
there can't be done. How interesting or it's that the instrumental
ones can't be continued? Let's try something that's
not an instrumental. Where's that skillshare
track that we made here? Put on your thinking
cap bits tell who's Sanan, how about the male one? You've got to rein,
that's better. Alright, Continue. Here it is. God, it's
still great Out. Let's give it some lyrics. Oh, look at that now
I can keep going. Continue. From which
point do I want to continue from? I don't know. Say can no longer hide. You're searching for a way to unlock the silence, you know? Let's try from that
chorus which was 17. 17. Why, what's wrong with that? That I do the dots wrong? Oh yeah, I did the dots wrong. You can't do anything wrong. You got to do it
exactly right, man. What a shame about
the instrumental. I guess those just
can't be continued yet. What do you
want to bet though? I'm going to say
definitively right now, you can't continue an
instrumental track. And then as soon as this weekend it'll be totally patched and you'll be able to do it
as much as you want. Maybe it's too hard to stitch two instrumental tracks together because no, it can't
be the length. You can make songs at 10
minutes long if you want, by stitching more and more
and more tracks together. It can't be that maybe
it's just too hard to put two different
instrumental tracks together or I've done
something silly, which is just as likely, man, this has taken a while. Part two is and part three, this is virgin two, where a part two is taking a lot more than a part one.
This seems the same. Be at night when the city sleep underneath the
Coke Street street. When London walk don't need it. Escination in line
sketching five and losing on my cheese. Neon dreams. Always neon dreams. If you use the AI to make your lyrics guarantee
it's going to be about neon dreams or about coffee.
Those are the two big ones. Oh well, in sunshine,
those are the big three. Anyway. Yes. So
continue from 17, I guess once we export it, we'll know what
that sounds like. The cut in of continuing
from 17. How interesting. Hm.
5. Adding MIDI & Vocals to AI: I have exported one of the instrumental tracks that
I made into Ableton is. Let's text the length. It is Oh, Pretty square on. Oh, yeah. It is exactly 2
minutes long. How interesting? Having a look at the track, it is 120 pm. It's exactly what I asked for. Yeah, good work on them. The one I chose is one that has an insanely long drum build up. And the reason I chose it is because it's going to allow
me to add my own layering. And it sounds nicer than if
a lot of instruments and layers existed in the track already for me to contend
with and to work over. What I'm going to
do is I'm going to make it a chord progression, And I'm going to
assume it did what I wanted and compose
it in C major. If it did, then my chord progression is
going to sound awesome. And if it didn't, then my red progression
will sound terrible. That's what it sounds like. Okay, so there is
where it's pretty, I'll probably just start like this and then drag
it how I want it. It's a nice point to get a
core progression started. So to do that, I've got the
track in audio stem here, and I've made a Midi
stem underneath. I'm going to mute the audio stem because that's going to
annoy the heck out of me. I'm going to press command shift to make myself a new Midi track. I'm explaining because not
everybody works in Midi. I know plenty of composers or producers that work
exclusively in audio, using samples and
stuff like that to mix up a lot of tracks and glue it altogether
in the way that they do. So Origins piano. It's my babe. I love it. Oh, maybe not. I think Serum is going
to be the way to go. Sorry, Origins just
changed my mind. Okay, for this, I'm going to increase the
unison as much as I can. Go degenerate a bit at a second. Oscillation probably give it
something foul like that. Maybe you can drag
it out that way. For those who don't
know what Serum is or how to use it right
here on skill share, I have a class, I think it's composing with a synthesizer that
covers serum in depth. Okay. I'm just going to make
just a standard C major, just a root to get going with and
probably just jump it up. Go for those who like scale turned on C
major, look at that. As long as I compose
within the yellow lines, I'm composing within the
presumed key of this song. I say presumed because that
hasn't been proven true yet. Okay. Make sure that's nice and low because
serum is super loud. That sounds horrendous
because me turning the unison on so much
has made it ridiculous. Hmm. Maybe this one
can have a few. Oh, God, that was
terrible. Cool. That's, that's, that's
how getting started. Then the next one can jump up, maybe to hear too far,
maybe a little jump. Cool. And then I'm just
copying and pasting it. Why did I make
such a small stem? Good question. I thought I would not have the
notes go this long. Maybe a little bit higher. M not bad thing. Maybe a little bit lower, but not quite as low. All
right. I quite like that. I'm going to put them
together, Command J, that's going to turn them into just one thing like
that and not two. Then the next one again. Again, maybe it will be good. Maybe this one could be
a little bit higher. Oh, so close. This one you know when you hear the chord that you hope it will be and then it's
not, and then you got a. That's exactly what I wanted. I love it. Maybe it
needs a bit of a reverb. I love my vahala, love it. I could be wrong. I may have to turn
it off. I'm going to give it just a little one. My favorite with
Bls Brass Black. Well, it's not my favorite. It's one of my favorites. Let's try it with the music. I'm going to command
J again, make it one. I think I wanted it here. No, I didn't. I forget. Whoops, I muted it. Well, I should have
listened to my original me, which thought the notes should be half as long
as they ended up being. It doesn't match
the music at all. With the notes being so long. Oh my gosh, so silly. I had that thought and
I just ignored it. Now this thing is super long. Mm hmm. It doesn't line up well. This is annoying. This is what it takes when it's like AI is like
someone I've worked with who's made a track
and now I'm having to like line up with
somebody else's track. Oh man, this is no good. Goes until here. Oh man. It's gonna be like composing for a movie where you need to change the lengths of everything to suit the film. Still a little bit early. Hmm, that's going
to be intriguing. Well, this initial part
is going to be a pain, but then once I got all
the note values right, all the lengths and stuff right, then it will be fun probably to about and then it's here. All right? I think I kind
of figured out how it's supposed to go, man. I think doing it with your vocals is going to be a lot better than
doing it with Midi. With Midi, I guess, unless you're just like really
patient, then it's fine because even then, it doesn't
feel 100% on the mark. All right. That should be right. More or less that one's not right. That's so cool, man, I was like smacked in the face with the introduction
of that baseline. I'm going to cut this cord here and I'm going to
insert a midi track. I'm going to put it down here. Wait, what have I done? Copy. I forgot to copy it. I think I'm going to put it down here. How have I got serum set up? I'm going to put serum
down here as well. Like this, I think
that's all I did. But I'm going to put
a longer reverb into it so that it kind of as a kind of a cooler transition from
one to the other like that. Oh my god. Oh my god.
Oh my god. Oh my God. I forgot about the volume. Gosh, serum is crazy
with the volume. It could be more ambitious right there with everything in the same stem
like it is here, it's so hard to see the
wave forms and the drums. Where is the lead line? Everything's all just one just
makes things challenging. Um, okay, so how does that track E? Okay, cool. There's a lot you could add to this, to be honest. But gosh, is it going to
take time because you can't see the individual
stems that goes into it? Definitely the smartest addition to AI generated tracks like
this is going to be vocals. You could put a really
cool sample over it. Either record it yourself or
get one from somewhere else. And it's going to be
a bit more successful than fiddling around
with Midi like this, trying to get it
exactly lined up. Although some people will
find it, it'll go fine. Yeah, it depends on what you
want to add and who you are. If you're a vocalist, just
get those vocals in there. Right? Here's some vocals. I've dragged in cinematics. Ascension, 100 BPM
C minor vocal. Not quite the right key, but still might fit anyway. First thing I want to do
is see the warp here. It's going to warp
it to our BPM. You would think that
that would be awesome, but more often than not, it distorts the vocals too much. We really should be
using vocals that suit our BPM naturally,
sudo key signature. But let's see how
it goes anyway. All right, it's way
too much vocal, let's just move that away. Maybe just the first part was the there you go. It might be better
with your own voice, With this type of track, you might want to just
have yourself being like get down in the club. I don't know what people say
in the tracks, but yeah, this definitively prove AI is ready for vocal
sampling from you, for your own vocals that it
wasn't before, but now it is. There's a lot more we can
do and that's exciting.
6. That's It! : That's it for now. Things are changing
all the time and version three of
pseudo AI is indeed only in its alpha version
there are going to be a lot of changes made once a proper
version three comes out. I anticipate a lot of changes. Again, this space is
changing constantly. I don't agree with people
that say that AI is taking or will take all of the jobs of music producers
and composers today. It's always going to
need the human touch. It's always going to
need people to guide it, to supplement what goes wrong. To give it prompts to move it
forward and move it around. Where as producers we're
not going anywhere. There may be a situation where producers aren't going to be needed who can't keep up with the technology and
how things keep changing. But I don't think
that we're going to become completely
superfluous. So I completely recommend not ignoring AI because
it's not going anywhere. But instead,
educating yourself as best as you can so that
you can keep up and you can remain employable as AI takes more and more of
their responsibility off us and changes our role, which is going
to become more and more supplementing what AI
can't do. And guiding AI. And adding to AI, that's what I believe the future of
music is going to involve. Thank you so much
for joining me. If you would like to take
part in other courses, I have lots more about
composing for film, and for making Lofi and EDM
tracks without AI, of course, and other stuff as well as well as music theory. If you enjoyed yourself
or if you didn't, I would really
appreciate if you would leave a review and
start a discussion. If you have anything to add, if you have any questions, I'd be more than happy
to chat with you also, I would love it
if you would make something in Sono AI yourself and link it for me in the Projects tab so
that I can check it out. I would love to do that and have a listen and see what
you're able to generate. Thank you so much
once again and I hope to see you again
here real soon.
Misici, Music Composer & Producer
