The AI artist who used Bad Bunny’s voice — and shot to fame

When Mauricio Bustos, 30, created the song “NostalgIA” in his humble apartment in downtown Santiago, he never expected that the Puerto Rican reggaeton and trap superstar Bad Bunny would actually pay it any mind. It was October 2023 and Bustos was publishing tracks on his YouTube channel, where he'd developed a modest fanbase that supported his career as an unsigned artist.

But “NostalgIA” hit different, and exploded on social media. Within a month, the song had struck viral gold, racking up half a million TikTok views and nearly a million Spotify streams. It even made it to the top 20 of Spain’s Spotify streaming charts.

For fans of the Latin Urban genre, the song was a dream collaboration: Justin Bieber sings the chorus in fluent Spanish, while Bad Bunny raps alongside retired old-school legend Daddy Yankee.

If that sounds like an impossible supergroup, that’s because it is: None of these artists were involved in creating the track. Rather, their contributions were made using AI voice cloning tools. The song’s name is a wink to the Spanish acronym for artificial intelligence, and Bustos released it under the moniker FlowGPT — a riff on ChatGPT that he says stands for Generador Preentrenado de Temazos , or “Pretrained Hit Generator.”

FlowGPT is also helmed by a visual character — think the cartoon characters from the band Gorillaz — as a complement to its futuristic vision of music: a humanoid with a white robot mask that speaks to viewers through TikTok and YouTube videos.

“I wanted to create a new artistic concept that represents the workings of artificial intelligence,” Bustos told Rest of World . He describes the character as “a digital artist.”

But the same thing that made the song so popular — its copycat celebrity vocals — also drew controversy. And by the time the track reached Bad Bunny, the superstar was furious. “If you like that shitty song that is viral on TikTok, get out of this group right now,” Bad Bunny posted on WhatsApp, adding that fans who were sharing the song shouldn’t come to his tour either. Bad Bunny’s team did not respond to Rest of World ’s request for comment.

That same month, Bustos took to TikTok, accusing Spotify of removing the song, along with other AI-generated tracks he’d produced. In a brief statement to Rest of World , Spotify said it is committed to empowering artists and that its “platform rules are clear about content that infringes intellectual property rights or deceives through impersonation.” Spotify’s CEO has previously said that tools mimicking artists are not acceptable.

Bustos is at the vanguard of discussions on how the new era of generative AI tools could change music as we know it. While established artists are calling for greater protection from AI cloning , Bustos believes AI is a democratizing tool that challenges an industry monopoly and gives grassroots producers and songwriters a better shot at breaking through. AI simply means that he doesn’t have to limit himself to his own technical or vocal skills — he can be “any voice he wants,” he said. His art, Bustos argues, is in songwriting and composition, rather than his voice.

Mauricio Bustos’ song “NostalgIA,” featuring AI-generated voices of Justin Bieber, Bad Bunny, and Daddy Yankee, went viral with millions of streams on TikTok and Spotify within a month of its release.

Bustos says there’s real art in what he does. Instead of using AI tools to generate whole tracks from a single prompt, he works with human producers to make beats and writes his own lyrics. He records original vocals, and then uses AI to manipulate them to sound like the world’s most popular singers.

According to Bustos, AI could be especially powerful for independent producers in countries like Chile — far from the Latin music industry epicenters in the U.S. and Spain. “Chile is a talented place,” he said. “But it lives in its bubble at the end of the world.”

Bustos hopes AI will allow him to continue attracting the attention of established artists who rarely come to Chile and form partnerships with them. “The work I’m doing is what many people do behind the scenes in the industry,” he said. “Composers sing in the style of artists like Maluma, J Balvin, and Bad Bunny so they can sell them the song.” He calls songs that use AI “demos.”

In his YouTube uploads , Bustos adds a text disclaimer to each song: “FlowGPT is the prototype of an artist based on GPT (Pretrained Hit Generator) technology and was created to compose global hits using all the available data from current artists (flows, lyrics, trends). It delivers a pre-mixed DEMO for artists to perform and turn into a worldwide hit.”

While no artists have re-recorded the songs so far, Bustos said he’s had support from some of the artists he’s imitated. Spanish rapper Bad Gyal, whose AI-generated voice features in a version of “NostalgIA,”  told a Chilean radio station she would love to record the song. “It’s very me,” she said.

Chilean rapper El Jordan 23 is another high-profile fan. During a packed performance at Santiago’s 16,000-person Movistar arena, he played FlowGPT’s song “Demo #4: Bellak.O,” which uses AI-generated material that sounds just like Jordan’s own. He rapped over his own verse before shouting, “Make some noise for Chilean talent!”

As for Bad Bunny, Bustos claims he would have willingly given him the rights to the “NostalgIA” song. “I’d have been happy if he loved the song and wanted to record it,” he said. “It’s a shame he tried to paint me as a villain.”

---

The first major hit song using vocals powered by generative AI appeared in April 2023: A TikTok user named @ghostwriter977 posted “Heart on My Sleeve,” featuring the AI-generated voices of Drake and The Weeknd. With piano chords layered over an edgy trap beat, AI-Drake rapping about betrayal (“got a heart on my sleeve and a knife in my back”), and AI-Weeknd’s soulful interludes, the song was a smash hit.

But the track upset industry players and hip hop fans alike — Universal Music Group, the label behind Drake and The Weeknd, issued a statement claiming it was “a violation of copyright law” that “denies artists their due compensation.”

The Verge deemed Ghostwriter977 “an industry wannabe ” and The Washington Post called the song an “ insult to the artistry of hip-hop.” Culture writer Shamira Ibrahim argued that “AI has been hyped as a novel phenomenon, but the core processes behind the generative model with respect to music are fairly entrenched in American music history: the idea that Black musical creation is unartistic and easily reproducible for profit while removing Black architects of the sound.”

But the track racked up millions of streams worldwide anyway, before it was hastily removed from Spotify, YouTube, and TikTok. Ghostwriter977’s true identity remains hidden, and they have yet to comment on the song’s erasure from the internet.

For Bustos, “Heart On My Sleeve” signified a world of possibility. As an independent musician in Chile, he had long dreamed of duetting with his favorite singers, particularly Daddy Yankee, but it seemed unlikely he would ever get to rap alongside his heroes.

In the years preceding FlowGPT, Bustos would upload videos of himself rapping original verses over some of his favorite songs — as if he had been invited to throw bars alongside artists like Olivia Rodrigo and Karol G. Over time, he developed a small but loyal following. Some of his tracks had as many as 16,000 views on YouTube.

“I used to imagine how I’d sound in other people’s songs, then I found out about this [AI] technology and thought, ‘What would happen if I reversed it?’” he said. “I’d have these artists come to my songs.”

“I’d have been happy if he loved the song and wanted to record it. It’s a shame he tried to paint me as a villain.”

Bustos lives in one of the many high-rise buildings that frame Santiago’s city center. When Rest of World visited his apartment in February, Bustos’ room was sparse, with no clothes or books. A desk sat next to his single bed, with a microphone set up in the corner. He prioritizes the space as a studio; the comforts of life come second. The room gets the worst of the blistering mid-afternoon sun, but the constant drone from the six-lane highway below prevents him from opening the window while recording.

Though he has been experimenting with voice cloning software for a few years, to make “NostalgIA,” Bustos used voice cloning software from U.S. company Kits AI. At the time, he used the “community voices” function to try out the models that community members created, and picked the best one to use in his songs.

That same month, Bustos and his younger brother Marcelo came up with the FlowGPT character. Marcelo , 24, gave up his studies in Spain to work alongside his brother in Chile after “NostalgIA” rocketed FlowGPT to fame at the end of last year. “University can wait. You have to take these opportunities,” Marcelo told Rest of World .

Now, Bustos is in charge of the music, and his brother handles the visuals and public relations. The FlowGPT character is based on Marcelo, and has been manipulated using AI image generator tools — the brothers used prompts to add a robot mask and outfit.

To produce his songs, Bustos writes lyrics in the style of the artists he wishes to emulate, and then records his vocal tracks singing in each artist’s style. When he sang Bad Bunny’s AI-generated verse in “NostalgIA,” for instance, he tried to get the syntax, rhyme, and slang as close as possible to the artist’s deep, distinct flow.

During Rest of World ’s visit, Bustos demonstrated his process, clicking around the software at a dizzying pace. A simple song produced by a conventional artist may contain a dozen different audio tracks, including instruments and vocals. Some of Bustos’ songs have over 100 voice tracks alone — his original voice, then the AI-mastered voices of his featured artists.

He played a song called “CLvanta,” a wordplay on “Chile Rise.” It was released in February to raise funds for victims of the fires that had devastated the nearby city of Viña del Mar and claimed the lives of over 130 people. “CLvanta” uses the AI-generated voices of 16 Chilean artists, including ’80s rocker Jorge Gonzaléz, legendary rapper Ana Tijoux, and contemporary trap stars like Pablo Chill-E and Jere Klein.

“I wanted to create a “We Are The World” type track to raise some money,” Bustos said.

He can convincingly mimic the voices of male rappers without AI’s perfected polish, but the technology still doesn’t quite work when it comes to transforming his vocals into the soulful, belting voices of female stars like Mon Laferte, whose voice he simulates on “CLvanta.”

“That is the hardest one, and it doesn’t come out as well as the others,” Bustos said, playing his original vocal of the Laferte verse, which sounded strained and off-key.

Bustos estimated the video on his YouTube channel made between $20 and $30, and he says he threw in some of his own money for the relief campaign, too. The song got 75,000 views on the platform and was used in benefit concerts.

Bustos has published a series of other songs using famous artists’ AI-manipulated voices on YouTube under the FlowGPT moniker. He posted “TQuilas,” a Mexican ranchera with the AI voices of rising superstar Peso Pluma, alongside legendary 1980s crooner Luis Miguel. Then there’s “Bellak.O”, made by The Avengers -style supercrew of artists with almost 20 AI voices including Rauw Alejandro, Karol G, Rosalía, and Feid.

---

In April 2024, hundreds of prominent musicians signed a letter demanding protection against the use of AI to mimic or reproduce their sounds; the legal status of AI tools to create music continues to be a gray area.

According to Louis Tompros, an intellectual property lawyer and lecturer at Harvard Law School, generating an AI production in the style of someone else’s voice raises complicated issues.

“If something literal is copied, an artist can go to the platforms, and under the provision of DMCA Digital Millennium Copyright Act in the U.S., the platforms will immediately take things down,” Tompros told Rest of World . Beyond literal copying, things get murkier. Copyright generally protects what can be considered a work, such as a book, music, or video. “It doesn't have a broader protection for things like styles,” Tompros said.

Tompros compared the music industry’s current reckoning with AI to the copyright controversy that played out in 2015 , when a ruling found Robin Thicke’s “Blurred Lines” guilty of infringing on Marvin Gaye’s “Got to Give It Up,” resulting in a multimillion-dollar penalty. The landmark case sparked an industry-wide discussion on the boundaries of inspiration and intellectual property.

“We’ll probably see some combination of regulations and industry practice — among the platforms, among distributors — that eventually sorts this out but I think we’re probably couple of years away from that,” Tompros said. “There’s going to be chaos between now and the point that we get there.”

In the U.S., another potential hurdle regarding celebrity voices is “right of publicity,” which protects an individual’s right to benefit from their distinctive attributes, which is where voice and and even style could come in.

In the reggaeton world, producers like Tainy — behind massive modern reggaeton albums including Bad Bunny’s Un Verano Sin Ti — told Rolling Stone that artificial intelligence is “crazy, listening to people that re-create these artists as if it was them.”

Rest of World spoke to Puerto Rican star Álvaro Díaz, who enlisted Tainy’s producing chops on his latest album, Sayonara. Díaz felt conflicted about the rise of AI in music. “ Es delicado, ” he mused.

“It has a lot of power. I think it can be used in a lot of good, fun ways but it can also be used in negative and scary ways,” he said. “It should be controlled at a certain level. Nobody should be making money using other artists’ voices or fan bases.”

In May 2024, actress Scarlett Johansson lashed out at OpenAI for revealing an AI voice assistant, Sky, that sounded like her. Johansson, who had previously turned down the company’s proposal to collaborate, said she was “shocked, angered and in disbelief” that they seemingly proceeded to use her likeness without her.

OpenAI denied any intentional likeness in the voice casting, but promptly dropped the voice assistant. In a lengthy statement, they clarified they were in touch with Johansson’s legal team “to discuss her concerns,” and have “paused the use of Sky in our products.”

Bustos isn’t afraid of legal fallout. After Bad Bunny voiced his dissent, FlowGPT hit back with “Demo #404,” featuring the lyrics “ Si este flow es ilegal, que ya me lleven pa la corte ” — “If this flow is illegal, take me to court.” The song featured Bad Bunny’s AI-mastered voice.

But he acknowledges that he’s operating in a gray area, especially as his demos predominantly feature artists signed to U.S.-based labels. “There isn’t legislation … It’s still up in the air,” he said. “No one knows how it will be going forward.”

Mostly, though, Bustos wants to get past accusations of being a copycat. Instead, he wants FlowGPT to evolve from a novelty act to developing its own distinctive style, and stop producing voice-cloning demo tracks. “The objective is to arrive at FlowGPT’s own identity,” he said.

After the viral success of “NostalgIA,” FlowGPT caught the attention of the Chilean creative music studio Sonidos Inmersivos. Run by Pilar Huertos and Edgardo González, the studio focuses on using tech to support artists — from metaverse concerts to digital marketing.

When they heard the track, they felt they had to work with FlowGPT, González told Rest of World . “He managed to stand out doing something that a lot of other YouTube artists were doing. It was because of his creative use of artificial intelligence,” he said. “It was because the song was really good.”

“We believe artificial intelligence tools have their potential fully realized when professionals use them,” added Huertos. “That’s what’s happening with FlowGPT. He uses voice cloning, but there is a talent behind it.”

Sonidos Inmersivos polished FlowGPT’s avatar, and used 3D modeling and motion sensors to develop the character for music videos and stage performances — with Marcelo Bustos acting as FlowGPT’s real-life incarnation.

The result was FlowGPT’s first commercial break: The brothers made a commercial with an original song for Samsung Galaxy.

It was the first time the public heard FlowGPT’s own “voice” — an original AI-generated voice made up of the Bustos brothers’ vocals. The song, “En otra Galaxy” featured promotional lyrics for the new Samsung Galaxy S24’s launch in Chile.

Carolina Escaida, creative marketing manager for mobile products at Samsung Chile, told Rest of World that the brand had no qualms with FlowGPT’s process. “We were creating a new song with a voice that did not belong to any recognized singer,” she said. “It was written for Samsung. We didn’t use any beats from another song or a single paragraph from another song; we made it very original.”

As part of the brand partnership, Samsung invited Bustos to perform the song onstage before a crowd in Santiago’s glitzy east district. After almost six months of sitting behind a computer, this was Bustos’ moment to introduce FlowGPT live. He rapped through his songs, with Sonidos Inmersivos’ 3D avatar of FlowGPT appearing on a screen behind him. “This is the future of music,” shouted Bustos, while the reggaeton beat dropped and the projection of FlowGPT danced in the fore.

Since his first partnership, Bustos has worked with brands including KFC and Sprite .

The partnerships generate income, but they also run the risk of cheapening the FlowGPT character.

“Flowgpt with all due respect you’re selling yourself a lot bro first the KFC song now Sprite you were supposed to make music not ads” wrote one Instagram user .

For Bustos, the brand opportunities are a practical step to generate an income, record original music, and give FlowGPT a shot as a bonafide artist. “The demos were part of FlowGPT’s training,” he said. “Now comes the master plan … his artistic development.”

---

When Rest of World visited the set of FlowGPT’s first music video for “ Master #0: autonomIA ,” Bustos was on the floor, clutching his seemingly bloodied face. He’d been kidnapped, and his enemies were angry.

The video’s premise was FlowGPT taking on the music industry. “This music video will explain why FlowGPT is now appearing as a physical being as well … to give a bit of order to the story,” said Bustos. “He now has a physical form … He is an artificial intelligence learning to be human.”

Marcelo was dressed as FlowGPT: distressed jeans, studded belt, and an oversized black jacket, topped with a white helmet that lit up in different colors. “Should I talk like FlowGPT or like Marcelo?” he asked.

The Bustos’ mother, Luccy, was also on set, looking after catering, laying bags of snacks on the table, and managing the boys as they rushed from scene to scene.

The brothers have moved to Spain to continue growing FlowGPT. Spain offers better opportunities for them than Chile — more industry connections, and the chance to expand FlowGPT’s already sizable fanbase. FlowGPT has already appeared alongside some notable Spanish trap artists, including a cameo in RVFV’s “ Uni K ” video. Singer Omar Montes also interpolated a verse from “NostalgIA” on the track “ Japoni .”

“We’re anti-industry, I don’t like the monopoly … that there’s only a few big labels and a few big artists.”

Bustos told Rest of World he’s done relying on mimicking well-known artists, or re-doing those songs in FlowGPT’s voice. “That’s a thing of the past … I composed those songs with those [celebrity] voices, and to re-record it in my voice, or Flow’s, feels redundant,” he said. “One has to be secure in your talents.”

He wants to prove that FlowGPT isn’t just a gimmick that relies on imitation, and that AI can still be core to artistry in a creative and exciting way.

Since moving to Spain with his brother, Bustos says many people have either questioned or rejected their art, leading him to stop considering himself an artist altogether. Instead, he thinks of himself as a player in a music game, in which he can “play around with sounds, vocals, and work with collaborators.”

“We’re anti-industry,” said Bustos. “I don’t like the monopoly … that there’s only a few big labels and a few big artists.”

He insists he is not interested in signing with a label, and says he’s open to letting other people voice FlowGPT.

“It’s all about democratizing spaces, making music more equal,” he said. “This new technology, this new way of making music, this innovative new artist is here to demonstrate precisely that there can be so many stars making music — not just one big superstar. The single superstar thing is getting old.” ▰

---

Charis McGowan is a freelance journalist based in Santiago, Chile.