Suno is an interesting tool for jumpstarting songs, but I would never publish a raw Suno track as finished. But the day is coming when “finished” tracks will be almost the push of a button. Remember, right now this is the WORST the technology will be.
Back to Suno stuff:
One usage pattern I’ve tried with some success is to:
1: enter my original lyrics into Suno
2: dice roll generative song/vocal segments with Suno
3: repeat until something cohesive and representative of my original idea takes shape
4: generate and download the “stitched” song
5: import into RipX to extract individual parts
6: Change vocal melodies and timings in RipX
7: export instrumental and vocal parts to stems
6: import stems into Ableton or DAW of choice
7: slice and arrange parts into target song structure
8: replace or remove all AI musical parts (make everything your own)
9: replace AI guide vocal with real vocals. Alternative: use service like audime to transform guide vocal or your own singing to “good” AI voice
10: finish producing to taste
Note that in this scenario, AI is mostly used as a song starter framework, to be replaced with higher quality elements and better human musical ideas.
Couldn’t you just avoid the first 4 steps if you sing your song? The rest of the steps will be just similar to any “urban artist” pipeline regarding voice processing.
Not to mention people using it, participating in its training, leaving feedback and using it FOR FREE while they are afraid they will lose their job to an to the very thing…ironic.
Sure, I could just sing the track myself, then use something like audime to transform that into a decent female vocal when those are required.
But what I’ve outlined is a process that could work for non-singers doing a solo track or even just a demo that requires vocals. I was effectively a non-singer for a while because I had a bad lingering flu and fatigue that turned my voice to doggy doo off and on for weeks on end, so part of my impetus for this process was to see what I could do minus a working singing voice.
I’ve only done this a couple of times and I’m back to from-scratch song production for my next batch of songs, but will likely use audime going forward for some female vocals as needed.
Yikes and whoa. Just did a prompt for shigs and gittles and, though it wasn’t even close to what I was asking for, it churned out a song that sounded more like Lana Del Rey than Lana Del Rey sounds. Personally I’d rather listen to LDR herself.
I honestly do not get who thinks these AI tracks are good to the extent that you’d actually want to listen to them as anything besides maybe background filler. They tick the superficial boxes of what constitutes a “piece of music”, but then execute in the most formulaic, boring and unsurprising way.
On a 10-scale, the best these tracks can reach is a 7, and that’s being generous and ignoring the vocals, mixing and mastering.
I guess the problem is that for a lot of people a 7 is the maximum extent they’re able to stretch their brain to understand interesting music. I’m being an elitist asshole, I know, but it seems like we need some elitism right now.
The thing to remember is the AI dev is at early early stage, it’s the Nokia 3310 of phone.
We can’t believe it will not be improved A LOT.
The tracks made right now are bad compared to human tracks, but imagine in 5 years, 10 years ?
The speed progress is so intense, you will not detect AI music in near future.
The technical aspects will improve. As long as the AI is based on machine learning from large datasets it will by definition be unable to go past a certain plateau of artistic expression.
In other words: it will be great at creating things that are based on the most common chord progressions, the most common melodic phrases, the most common instrumentation. And the most established, clearly defined genres are the easiest to generate content for.
Sure, you can force it to make genre hybrids like make reggae for me with heavy metal electric guitars. And it will make another song that’s maximally a “7”, and it will be a funny joke for those people who laugh at “7” level jokes.
No, you don’t understand. The AI’s are based on large dataset models. It will only learn Leon’s Control-All tricks if everyone else starts using them too (or Leon uses them a lot and becomes very popular).
