Minimax Music - A song generator for musicians

[Kenmac]

Just giving everybody the heads up on a new (to me) song generating website that's different than Udio or Suno and others called Minimax Music. The difference with this one, as you'll see in the video demo by YouTuber Bob Doyle, is that you have a lot more control over aspects of your song. For example you can tell it what key you want the song to be in, the number of bars for different sections, 16, 24, etc. What instruments you want in the song and even how you want the lead vocalist to sound. The only thing missing is being able to tell it what notes the lead vocalist should sing but I wouldn't be surprised if something like that happens in a future update. Go to 9:52 in the video for the music portion. (I tried to paste it in from YouTube but it gave me the full video.)



The good news is right now it's free to use so check it out here:
https://www.minimax.io/audio/music

[VOODOO U]

As far as generating songs it has features that Tunee lacks so that's a huge plus. But I'm not sure about #5 in the terms:

5. Markings and Notices on Output Content You acknowledge and agree that, to comply with legal requirements or clarify rights and restrictions associated with Output Content, we may embed identifiers or watermarks in such content. When you view, export, or download Output Content, we may display prominent notices informing you of the applicable usage restrictions. Unless otherwise authorized, you may not alter or remove these identifiers. You are also responsible for clearly labeling AI-generated Output Content in your use to avoid misleading others. You assume full liability for any consequences resulting from failure to provide appropriate notice.

I don't like the idea of possibly having identifiers in audio Output Content to acknowledge restrictions.
Also, is it mandatory to label A.I. usage if I only use partially generated output content (ie a vocal stem)?
What if I take a guitar stem and play it myself as a new recording?

[Kenmac]

Yeah, I can see why that would be concerning but also notice that they use the word "may", so they might not embed identifiers or watermarks or display the notices about usage restrictions but if this bothers people they should stick with another music generator. I can't answer your question about the guitar stem. Maybe get in touch with Minimax to clarify.

[BONES]

I don't like the idea of possibly having identifiers in audio Output Content to acknowledge restrictions.
Also, is it mandatory to label A.I. usage if I only use partially generated output content (ie a vocal stem)?
What if I take a guitar stem and play it myself as a new recording?

Don't overthink it. I'm sure that's there to cover their arses if someone uses it to impersonate Drake or Taylor Swift or someone. They aren't going to come after you if it's a legitimate creative endeavour. Adobe offer the same thing, but they see it as a service to creators, not a burden, so you can keep track of your own output. Any identifiers would have to be in the metadata, probably buried where it's harder to get at, so if you use a stem in a bigger project in your DAW, the metadata won't be there in the rendered output, so you'll have nothing to worry about.

Personally, I don't see any advantage in being able to tell the AI the key or how many bars or other things like that, it's just putting too many restrictions on its creativity. We've always got our best results from giving it its head and letting it rip. Our prompts focus on the emotional side of the piece, how we want it to feel - you know, that stuff everyone keeps saying AI is no good at. We don't bother with the technical shit. Unless you are doing a jingle for a TV commercial or something similar, I don't see what relevance it has. Even if you were looking to create a new part for something you're working on, I still think it might be more fruitful to adapt the existing piece to what the AI spits out. After all, once it's given you a file, you can edit it however you want to - separate the stems, copy/paste parts, etc. - until you have something that works the way you want it to. That's what we've been doing for the last 7 months.

[VOODOO U]

I can't answer your question about the guitar stem. Maybe get in touch with Minimax to clarify.

I'd bet I get a quicker, detailed response going to chatGPT. Just the Minimax account sign up itself is a hassle for those who don't want to use a Google account. I have a Google account but it's the principle of a company forcing people to a certain direction for use of their services that bothers me.

Any identifiers would have to be in the metadata, probably buried where it's harder to get at, so if you use a stem in a bigger project in your DAW, the metadata won't be there in the rendered output, so you'll have nothing to worry about.

Ah yeah didn't think of that one. Problem solved.

Personally, I don't see any advantage in being able to tell the AI the key or how many bars or other things like that, it's just putting too many restrictions on its creativity

With the results I get, more so than not, the output contains shit I like to make changes to. I want control over BPM, length of song in minutes, vocal delivery and avoiding things like solos and long intros.

Our prompts focus on the emotional side of the piece, how we want it to feel - you know, that stuff everyone keeps saying AI is no good at.

It's been up and down in my experience particularly with vocal delivery. I like raw and dirty female vocals and I keep getting shit vibrato and Shania Twain trying to sing punk-ish. I found the trick is in the lyrics. If I give it serious lyrics I get shitty results. If I give it lyrics about ketchup on a cow's ass I get better results.

Even if you were looking to create a new part for something you're working on, I still think it might be more fruitful to adapt the existing piece to what the AI spits out

Thing is more so than not I really like the particular sound characteristics the A.I. outputs better than the sounds I could get with my own gear.The guitar tones Tunee outputs outshines what I could get. It's all shot to shit if I want to make any changes even if they're minor ones. I can't tell A.I. to redo the same song exactly the same way but add a specific bridge. That's an issue if I want to keep the actual output the A.I. produces.

[BONES]

With the results I get, more so than not, the output contains shit I like to make changes to. I want control over BPM, length of song in minutes, vocal delivery and avoiding things like solos and long intros.

Really? We've never had any solos and I have had to add intros to half the songs we've got out of Tunee. The things it spits out are always too short so I just carve it up and add a few bars here, a few bars there, repeat the chorus a couple of extra times, etc., until I'm happy with it. Weirdly, if I'm not paying attention every song will end up somewhere between 4:20 and 4:40 long. That seems to be the range when it starts to feel "right" to me.

It's been up and down in my experience particularly with vocal delivery. I like raw and dirty female vocals and I keep getting shit vibrato and Shania Twain trying to sing punk-ish. I found the trick is in the lyrics. If I give it serious lyrics I get shitty results. If I give it lyrics about ketchup on a cow's ass I get better results.

Interesting. Mostly I love the vox Tunee spits out. It's in a very different style to mine, more Nine Inch Nails than Front 242, but it doesn't matter because I ruin them all with my own dulcet tones.

It's all shot to shit if I want to make any changes even if they're minor ones. I can't tell A.I. to redo the same song exactly the same way but add a specific bridge. That's an issue if I want to keep the actual output the A.I. produces.

Yep, we find exactly the same thing. For the one song that I got out of Tunee for the album, I did 18 versions but they just got further and further away from anything we could use, so I ended up using the first thing it spat out. It's turned out OK, although I'm not happy with my vox so we'll have to re-record this week.

[VOODOO U]

Really? We've never had any solos and I have had to add intros to half the songs we've got out of Tunee. The things it spits out are always too short so I just carve it up and add a few bars here, a few bars there, repeat the chorus a couple of extra times, etc., until I'm happy with it.

Well we're producing different styles of music. I aim for the old skool punk circa late 70s style which Tunee gathers needs a solo here and there. No issue until there's a damn solo in every damn song or at least it's a solo I can't stand to listen to. With most industrial styled music anything under 5 minutes is too short whereas with punk styled influences anything over 5 seconds is 5 seconds too long.

Interesting. Mostly I love the vox Tunee spits out. It's in a very different style to mine, more Nine Inch Nails than Front 242, but it doesn't matter because I ruin them all with my own dulcet tones.

Ruin them all is right. If I want sweet I'll grab me a snickers bar. Now there's a song title.