Whilst you click on at the Spotify profile of Clever Band System you are going to see a picture of 3 younger males staring moodily again into the digicam. Their profile confirms that they’re a “British band”, “influenced by the post-punk scene” and looking to seize the spirit of bands like The Remedy “while carving out their own unique sound”. Whilst you concentrate to their song you could be reminded of Pleasure Department’s Ian Curtis.
In the event you dig somewhat deeper and examine them on their file label’s web page you are going to to find that Cameron is the lead singer and his musical tastes have been formed via the live shows he attended at Nottingham’s Rock Town nightclub. Tyler, the drummer, was once certainly impressed via The Remedy, in addition to U2, and The Smiths, whilst guitarist, Antonio, blends his Italian mom’s love of vintage Italian folks songs along with his British father’s hobby for The Beatles and The Rolling Stones.
What those profiles don’t say is that Clever Band System isn’t actual, no less than now not within the human sense. And I will have to know, as a result of I created them.
I used a spread of Generative Synthetic Intelligence (GenAI) gear, in addition to my abilities as a certified songwriter and sound engineer to make their debut album, Welcome to NTU, and I launched it on my devoted AI file label, XRMeta Information in Might 2025.
It’s possible you’ll ask why an independently freeing singer-songwriter and song manufacturer like me would create a synthetic band. In addition to being a musician, I’m an educational with a background in pc science, sporting out analysis about how GenAI can be utilized for song.
I had reservations about those gear and the way they may impact me as a musician. I had heard about quite a lot of AI controversies like “fake” Drake, and artists like Grimes embracing GenAI in 2023. So, I used to be additionally intrigued via the probabilities.
Over 100 million other folks have attempted Suno, an AI song technology platform that may create songs with vocals and instrumentation from easy textual content activates. Greater than 100 million tracks had been created the use of the Mubert API, which permits streaming to platforms like YouTube, TikTok, Twitch and Instagram; and in line with Deezer 28% of launched song is absolutely AI-generated.
It was once time for me to analyze what those gear may do. That is the tale of ways I experimented with GenAI and was once remodeled from a dance artist to a post-punk cushy rock band.
GenAI has modified the whole lot
In my early days of songwriting one of the crucial first items of apparatus I purchased was once a Panasonic RQ-2745, a small slender transportable cassette tape recorder that allowed me to file tough drafts of vocals on an audio cassette tape.
When reasonable merchandise just like the Sony cfs-w30 boombox started to include double cassette decks, I may overdub songs and upload choruses or tools like flute or guitar at house. If I sought after a high quality recording, I needed to e book a recording studio. I was a professional at splicing tape to take away vocal portions from the tape recording or to mend tape jams.
Reducing and taping, was slicing and pasting as I experimented with the very early unfastened virtual song sequencers that have been integrated on a disk I discovered at the duvet of a PC mag. I felt liberated when sequencers like Cubase, Professional Equipment, and Good judgment allowed top quality recordings to be produced at house. This, together with the numerous relief in the price of studio apparatus, resulted in the emergence of the bed room manufacturer and the proliferation of the 808 sound. This deep, booming, bassline can also be heard in hits like It’s Difficult via RUN DMC, Emergency Room via Rihanna, and Inebriated in Love via Beyoncé.
Virtual distribution and social media then cleared the path for self-releasing unbiased artists like me to keep up a correspondence at once with enthusiasts, promote song, and bypass file labels.
An instance of an early transportable tape cassette recorder.
Shutterstock/Dmitry Naumov
But right through all of those adjustments musicians nonetheless wanted the talents and information to create their songs. Like many musicians I honed my abilities over a number of years, studying to play the guitar, flute and piano, and creating sound engineering abilities. Even if AI powered gear started to be integrated into virtual audio workstations, a musician’s ability and information was once nonetheless wanted to make use of those gear successfully.
With the ability to create song from textual content activates modified this.
Now not because the advent of song streaming products and services within the past due Nineties has there been this kind of dramatic shift in song composition and listening applied sciences. Now non-musicians can create studio high quality song in mins with out the intensive coaching that I had, and with no need to shop for tools or studio apparatus.
Now somebody can do that. It was once time for me to be told what those gear may do.
I normally produce RnB/neo soul, nu-jazz and dance song, even supposing I will write songs for a couple of genres of song. For the experiment, I sought after to check out a style that I don’t typically produce song for.
The Insights segment is dedicated to high quality longform journalism. Our editors paintings with teachers from many alternative backgrounds who’re tackling a variety of societal and medical demanding situations.
I examined about 60 other GenAI gear and platforms. Those integrated standalone gear that concentrate on one job, like MIDI technology (musical knowledge that may be performed again on a keyboard or song sequencer). I additionally attempted AI song studios. Those platforms have consumer pleasant interfaces that mix a spread of AI gear to improve lyric, song, symbol and video advent.
Suno and Udio have been two of the most efficient platforms. They are able to generate songs with complicated vocal melodies and harmonies throughout a spread of genres, with the most efficient outputs being tough to differentiate from what human musicians can create. Each Telisha “Nikki” Jones and song tycoon Timbaland are stated to have used Suno to create song for his or her AI-generated artists.
In June 2025, Timbaland introduced the signing of his AI artist TaTa to his devoted AI file label, Degree 0. In September 2025 Jones was once reported to have signed a US$3 million (about £2.3 million) take care of Hallwood Media for her AI-generated artist Xania Monet.
On the time of my experiment in March/April 2025, each Suno and Udio had problems, similar to silence gaps, pace adjustments, inconsistent vocal high quality, and diversifications in style. Infrequently the voice may exchange throughout the tune. There was once restricted keep watch over on the subject of modifying, and the audio high quality may range inside a unmarried observe or throughout a chain of songs.
After making an attempt a number of GenAI song platforms I made up our minds to make use of Udio because of the standard of its output and its beneficial phrases and stipulations at the moment. Taking inspiration from pop-rock and post-punk bands like Pleasure Department and The Remedy, I began the adventure in opposition to growing a brand new personality.
The use of GenAI to supply one or two excellent songs was once fairly easy. Generating an album of 14 songs that sounded as though they have been performed via the similar band was once more difficult, in particular producing the similar male voice and musical taste for every tune.
The songs have been both a ways too very similar to every different or had different problems such because the voice converting, or the tools sounding too other. A cautious concentrate to the songs in Spread out via the AI artist Xania Monet will expose equivalent inconsistencies. As an example, you’ll be able to pay attention a distinction within the voice this is generated for the primary tune, This Aint No Tryout, in comparison to Again When Love Was once Actual.
GenAI can’t write (first rate) lyrics
My first job was once to create the lyrics. I generated about 1,000 songs the use of Udio and located repeated phrases and words within the lyrics like “neon”, “whisper”, and “we are, we are, we are”, showing each inside and around the two consumer accounts I created. Issues like darkness, shadows, and lightweight have been additionally repeated throughout the lyrics for a vital choice of songs.
GenAI simply couldn’t write lyrics with the complexity or playfulness I wished, so I selected to write down the lyrics for the album myself and used a semi-autobiographical narrative. This allowed me to take care of a tale around the album; from arriving at Nottingham Trent College and settling into pupil lodging, to experiencing college existence, graduating and leaving.
I may interweave present affairs just like the closure of Nottingham’s Victoria Centre Marketplace within the tune Good-bye Vicky Marketplace. I integrated strains that referenced Nottingham’s ancient figures like Alan Sillitoe, who wrote The Loneliness of the Lengthy Distance Runner, and the creator D.H. Lawrence, within the tune, Books.
After writing the lyrics I generated the song. There have been problems with urged adherence. I examined activates of various lengths. In some instances, activates have been in part or wholly neglected. I may write a urged requesting one style and a distinct style could be produced.
There have been additionally problems with the substitute voice announcing one of the lyrics. As an example, it might now not pronounce “NTU” or “Sillitoe” and I needed to rewrite one of the lyrics phonetically or edit the audio to get the proper pronunciation for positive phrases.
I depended on my sound engineering abilities; extending the outputs, modifying, blending, remixing, and manually recording vocals in Cubase to succeed in a coherent ultimate combine. This took a vital period of time. In reality, modifying the Udio outputs took such a lot time it could had been more straightforward to recreate the song myself. I will write a tune in ten mins, and I every now and then file myself freestyling lyrics for a whole tune at once in Cubase, so this was once irritating.
I encountered equivalent problems with urged adherence when producing photographs and video. When the use of Kling AI to create photographs of the band individuals, I adopted its urged engineering information. Then again, I needed to generate masses of pictures and edit them with exterior gear to succeed in the general band pictures.
Producing video was once similarly difficult. One method to create a video is to add a photograph, which turns into the primary body. The remainder of the video is generated in line with the urged. Then again, after I uploaded Cameron’s profile symbol to Kling AI, the preliminary frames of the ten-second video resembled him. However via the tip of the video, Cameron regularly morphed into any person else, and this came about often when producing video.
Activates for digicam directions, similar to zoom and pan, have been often neglected. I additionally needed to edit out scenes with different issues, similar to the illusion of additional palms or an extra leg at the band individuals.
All this wasn’t reasonable both. With 8,000 Kling AI credit at a price of US$64.99 (about £50), I may generate about 40 ten-second movies, however many have been unusable.
Song technology is less expensive. Paying between US$24 and US$30 (kind of £18-£24) for a per thirty days subscription may permit a consumer to create between 2,000 and three,000 songs, relying on how the “credits” are used. I used to be very stunned to find how briefly those tune credit can also be fed on. Each and every error or tune that didn’t go well with my style nonetheless price credit.
In the end, after producing 1000’s of songs, masses of pictures and video, the use of gear like Duck.ai to create the band’s biographies, and spending many hours modifying the outputs; Cameron, Tyler and Antonio started to emerge because the band.
AI-generated band Clever Band System: Antonio, Cameron, and Tyler.
Priscilla Angelique-Web page
One thing sudden came about
I’ve all the time been hooked in to growing my very own song. Up to I really like writing songs, the deficient royalty payouts I used to be receiving had grow to be disheartening. A tune I recorded in 2001 and launched in 2011 referred to as Most effective Heaven Can Examine was once streamed about 1 million instances in France right through 2024 however I best gained about £21 in royalties.
Previous to streaming, had my tune been downloaded via simply 10,000 other folks, I’d had been paid about £6,900 (69p in step with obtain). Artists like Kate Nash have raised considerations concerning the deficient royalty payouts to musicians, bringing up her £500,000 payout for over 100 million performs of her tune “Foundations”.
However as I created the band’s album one thing sudden began to occur. I started to experience growing song once more. The frustrations with the use of GenAI was once balanced via surprise and interest.
From time to time Udio was once ready to generate vocals that have been so reasonable I may infrequently consider they have been created via an AI fashion. There have been moments after I laughed, when I used to be truly moved, or even had chills after I heard one of the songs.
Lyrics that when lay dormant in a couple of lever arch recordsdata on my bookshelf started to seek out new existence thru those generative gear, permitting me to hastily check them throughout a couple of genres.
I made up our minds to take this experiment additional.
After in moderation deciding on a collection of songs I had written a few years in the past, I created a brand new personality, Jake Davy Smith. For his 14 observe album, referred to as I’ll Be Proper Right here, which was once launched on November 22, 2025, I used Suno’s v5 fashion to generate studio high quality song that matched my unique imaginative and prescient.
Suno’s intensive modifying gear permit customers to add vocals, create a canopy tune, and edit the song, lyrics, or voice with higher precision than their previous fashions. This helped me just about recreate my unique songs. The observe Calling is an instance of a rock ballad I wrote years in the past, recorded and didn’t unlock.
Conflicting feelings
Reflecting in this experiment, I discovered myself with conflicting perspectives about the use of GenAI. Those gear are rapid and inexpensive (in some instances, utterly unfastened). They are able to produce quick effects. I now have gear that I will use to briefly reimagine my previous songs.
I will use a couple of personas to carry my lyrics to existence. I’m Priscilla Angelique. I’m Clever Band System. I’m Jake Davy Smith. I’m Moombahtman 25, a male African American moombahton artist who combines hip hop with Latin American beats, and I’ve many extra personas.
I’m a “multiple persona musician” or MPM, a time period I’ve created to outline my new musical identification. Musicians having modify egos isn’t new, however GenAI has utterly modified how that is executed.
Then again, there’s some other aspect to this. Human musicians at the moment are having to compete with algorithms able to generating top quality song at scale – in addition to with every different.
Those gear are bettering hastily, and the problems I skilled when the use of Udio to create the album for Clever Band System in March/April 2025 have already been addressed in Suno’s v5 fashion. It’s now more straightforward to create a personality with a constant voice. Customers can add their very own songs and likewise create duvet variations in their songs.
Growing the album for Clever Band System took about one month and there have been a couple of problems with looking to create persistently sounding top quality AI-generated songs. I spent hours reviewing 1000’s of outputs after which extra time modifying the general set of curated songs in Cubase.
My revel in was once very other after I created the album for Jake Davy Smith. I used lyrics I had already written, generated between 5 and 20 variations of every tune, and spent a ways much less time modifying them. The method was once sooner, then again, there have been nonetheless some problems. Adjustments in Jake’s voice took place, even though they have been much less common and more straightforward to right kind. There have been additionally issues of pronunciation, however I may now briefly regenerate the audio. In essence, what had up to now taken a month now took just a week.
Moral problems and information assortment
But underneath this lies an extra inner war associated with the information used to coach those AI fashions or, as song journalist Richard Smirke describes it, “the largest IP theft in human history”. It’s this factor that has made a era that should had been celebrated as one of the crucial greatest technological achievements in a long time, some of the contested as an alternative.
Chatbots like ChatGPT, estimated to have 1 billion customers international, had been described via the linguist and activist Noam Chomsky as each “marvels of machine learning” and the “banality of evil”. Symbol turbines like OpenAI’s DALL-E have additionally come underneath hearth. Critics like Ted Chiang problem whether or not AI could make artwork and different commentators have criticised the loss of cultural range in symbol technology.
Jake Davy Smith AKA Priscilla Angelique-Web page.
Priscilla Angelique-Web page
So GenAI is right here to stick, and musicians will want to adapt. Library song, background song, and song for social media or movie can simply be created with AI. Then again, there are dangers. The chance that equivalent song could also be generated for different customers; the chance that any uploaded songs could also be used for coaching knowledge. Then there’s the chance that those gear might inadvertently generate one thing that breaches any person else’s IP.
A technique for musicians to securely use GenAI is via coaching fashions the use of their very own knowledge, as YACHT did after they used their again catalogue of songs as coaching knowledge for a brand new album. On this method musicians could have complete keep watch over over the outputs. That is one thing I can be exploring for the following degree of my analysis.
What AI can’t do
My transformation has been the rest however simple. It’s been marked via the deep frustration I encountered when to begin with the use of those gear, an ongoing war about how those gear are educated, and moments of authentic amazement. The albums I created could also be imperfect, however they’re a transparent departure from my same old taste and display how GenAI can improve musical creativity.
Priscilla Angelique-Web page aka Priscilla Angelique, dressed in an AI-generated jacket and blouse.
Priscilla Angelique-Web page
Financially, the albums are not likely to recoup the price of growing them, as unbiased musicians might want masses of hundreds of thousands of streams to earn a good source of revenue from song. Even a couple of million streams of the songs will slightly duvet the quite a lot of charges for song, symbol and video technology of round £140. Products, licensing, sync offers and different earnings streams will most likely stay essential assets of source of revenue for musicians, whether or not they’re human or AI-generated.
At the felony aspect, one imaginable method ahead is for AI corporations to make open-source variations in their fashions freely to be had for offline use. Some have already got, however for those who haven’t, it sort of feels truthful that if they’ve used our knowledge to construct those techniques, they will have to permit broader get entry to to the fashions themselves.
New applied sciences may exchange how song is produced. We’ve got long gone from clapping to drumming, and from the use of drum machines in recording studios to producing “new” sounds with AI. But now that I’ve finished those experiments, I realise that something stays the similar.
Whether or not I’m slicing tape the use of scissors, slicing and pasting in a sequencer like Cubase, or regenerating portions in an AI song studio like Suno, human creativity remains to be an very important a part of the method. The use of GenAI was once transformative, but it was once my inventive selections that formed the songs, the albums, the avatars for my personas, their biographies, and the whole imaginative and prescient. That is one thing that AI can not do – no less than, now not for now.
For you: extra from our Insights collection:
