Here’s a creative AI irony for you:
The professional music industry is the most accepting creative group toward AI, yet AI music is not nearly as popular in the AI community as other mediums.
We should change that. To do that, we must spread our knowledge about AI music and how to create it.
Today, we're taking a look at Suno which is currently the best AI app for creating music with lyrics.
(Note: I use the web version of Suno exclusively, so the Discord version may be different.)
Lyrics
First thing’s first: we need our words!
You can use Suno’s native generator for this, but I find it produces very same-y lyrics. I’ve generated all my song’s lyrics thus far with ChatGPT but I wouldn’t be surprised if Claude is also very good at this.
What Suno's generator does do that others don't is include echos and calls such as (ohh-yeah). You can add these manually or instruct the LLM to specifically include these. In fact, you can use Suno's lyric generator to generate examples for the LLM to follow.
Whatever LLM you use (or don’t if you create your own!) here are some areas you want to be aware of if you want a song that is consistent.
Meta Tags
Meta tags are the words in brackets that aren’t said during the song. They can include things like [Verse], [Chorus], [Prechorus], [Bridge], [Outro], and [Intro].
Suno is fairly good at understanding verse, chorus, prechorus, and bridge, but not so good at understanding outro and intro. I’ve tried using tags like [Instrumental Intro] with very limited success – perhaps 1 song takes it into account for every 2-4 songs generated.
I do not recommend using numbers such as [Verse 1] and [Verse 2] in meta tags since these can result in the segments sounding very different.
Verse to Chorus Length
Suno’s initial generation maxes out at 1 minute and 20 seconds while continuations max out at 1 minute. You want to be able to fit at least a verse and chorus within that 1 minute.
Why is this important? Suno does not seem to look at the whole song when generating a continuation: it only looks at the clip you’re continuing from. That means, if you want the verses and choruses to have the same sound, Suno needs to have that data directly in the clip it’s continuing from. If it doesn’t have a verse/chorus to reference, the new verse/chorus will very likely sound different from previous ones.
If you’re using a bridge instead of a final verse, just make sure it still has the data for the final chorus.
It’s difficult to estimate how much will fit into one minute since much depends on pacing, but I’ve found alot of success with verses and choruses that are each four-lines long.
Syllables Per Line
You want to give Suno rhythms it can easily use and repeat. I suggest giving it around the same amount of syllables in the same lines for each section type.
Suno likes 4/4 rhythms* but that doesn’t mean every line needs to be four or eight syllables – it’s good at making use of holds, half-notes, and pauses to fill in odd syllable counts. If you do have longer lines, commas are a good way to direct Suno where to break the line.
(* I have never tried creating a waltz. If anyone has, let me know – I’m curious if it works!)
You can ask ChatGPT for a specific number of syllables per line when initially generating, but I have found it prefers odd numbers to evens, so keep that in mind.
I recommend reading your lyrics and seeing if you can hear a rhythm. If you can’t naturally hear one, Suno will likely have difficulty too. If you are having a song that Suno is generating very inconsistently, try giving it to ChatGPT or Claude and asking for help fixing the rhythm.
Style
There is a character limit on the style prompt box, so keep your prompt limited to brief terms or descriptions separated by commas.
Suno recently added a “random style” which generates a two or three word style. I haven’t played much with it yet, but this might be a good starting point. Here’s some things to think about:
Vocals: Suno is generally good with requests for either male or female vocals. It’s less good with adjectives like “layered” but they’re worth trying.
Pace: I haven’t had much luck with asking for specific BPM, but I recommend adding a term that denotes pace. Descriptions such as “high-energy” or “relaxed” will do.
Instruments: Suno is not good at adding or emphasizing specific instruments, but requesting them can change the sound of the song nonetheless.
Generations
Each time you hit “create,” Suno will generate two songs with each costing five credits a piece. Very rarely Suno will accidentally generate a very short clip of 2-5 seconds, but the system is good about catching these and refunding you the credits for that song.
Occasionally Suno will generate an instrumental even though there’s lyrics, or generate gibberish instead of the specified lyrics. The system can’t catch these, but they happen perhaps one in ten generations.
Suno has a very generous free plan: 50 free credits a day. These do not roll over to the next day, so be sure to use them each day. Songs created with the free plan are under a non-commercial license. If you want to use the music for commercial purposes, you will need to subscribe.
Simple vs. Custom Prompts
Since I already spoke about lyrics, you can guess that I’d recommend using Custom Mode, which has a toggle at the top of the box. Not only does this give you access to the lyric box, but it also allows you to name your song.
Initial Generation
Your first generation is your most important one since it sets the stage for subsequent generations. You want to look for a good rhythm and melody, but above all else, quality. Subsequent continuations are fairly likely to diminish in quality, so you want a strong foundation for your song to sound as good as possible.
Right now, Suno is cannot produce a very “full” sound with alot of background instruments in a high quality. Similarly alot of bass in the background will also reduce quality. The simpler your song is right now, the better.
Continuing Clips
Once you have a generation you like, click the three dots on the song and select “Continue from this Clip.” Once you have continued at least once, you have the option to “Get Whole Song.” This option does not cost credits.
It used to be that Suno seemed to generate from a Whole Song better than a series of clips, but that does not seem to be the case anymore.
Once you have a few generations, your dashboard may start to get confusing. If you thumbs-down a clip, it will disappear from the dashboard, helping you keep it cleaner. You can find these clips again in your library.
Similarly, you may want to thumbs up the clips you plan to use or that you are impressed by. You can filter your library by liked clips, letting you find these easier later.
Cut-Off Clips
It is very likely you will generate a clip with the last word or more cut off. If you want to continue from these clips, start with the next word, not the one that was cut off. The vast majority of the time, the word will finish with no problem in the continuation.
There can also be a problem with clips ending a song before you want it to. This seems to occur most often when the last couple seconds of the previous clip sounds like it’s wrapping things up – at least to Suno. If you are having a problem with these ending clips, you may have to regenerate the previous clip so Suno has a better point to continue from.
Conclusion
Suno is the best AI music app right now for songs with lyrics. That combined with a very generous free plan means you can experiment with the app every day and put together some amazing songs.
Now get out there and create some music!
Kommentare