AI Music

Over the summer, I experimented with a few services that create music and vocals when you supply lyrics to a generative AI. I tried Sono, but had poor results. Eventually I found a service called Riffusion, which has since been renamed to Producer AI. I found it entertaining as I experimented with different aspects of trying to make it do things that weren’t quite normal, such as comedic monologues on dry humor, life of domesticated punk/goth people, tongue twisters, nonsensical lyrics, and serious sounding songs in latin about comedic events such as a dog farting at the dinner table of senators.

Riffusion / Producer AI: The Bard of Bizarre

Eventually, I started putting together a tragicomedy concept album about an office workers day where just about every object around him is personified and out to get him in one way or another. Each song was done in a different genre to help express the mood throughout the day. The lyrics went through quite a few iterations, ensuring that everything had continuity and that each song referenced things that happened in prior songs, and that some objects such as a cracked phone continued to be referenced throughout each song.

Riffusion has a licensing scheme where you have rights to sell the music – but only if you pay for a subscription. After I got comfortable with the system and tested the ability to create a concept album, I looked into the costs to do so, and what the conditions were. The first issue that I ran into was that anything made using the free version did not upgrade into a licensed version once you buy into a subscription. So I contacted the company to inquire about existing songs. The reply was that they could review up to four songs and then charge for the licensing. Mind you, a subscription was $432 per year.

So in the end, I had to re-run all of my lyrics to generate new songs that I could license – which honestly, I had to do anyway as the original lyrics were fairly raw with many continuity issues and lacking in personal details. In the end, I have two separate concept albums – one I can sell, and another I can only share. I wish I could get the last song, “Too Tired to Sleep, Somehow” to have the same sound as the original song I had generated. The same goes for “The Couch Pulls Me In”.

Cubicle 13

#	Time	Title	Description
1	7:10 AM	Time Is a Flat Liar	Alarm clocks lie. The calendar gaslights. A dramatic synth-pop opener to the cursed day.
2	7:24 AM	The Cursed Brewer’s Soliloquy	Office coffee doesn’t just betray taste – it betrays your soul.
3	8:11 AM	Buttoned Up, Butt Taken Hostage	Wrangling pants before a morning meeting becomes an existential battle.
4	9:03 AM	Locked Out, Keys Laughing	He forgot his keys. They remember everything.
5	9:27 AM	Sent to Print. Never Seen Again.	Industrial Rock meets printer possession. Karen’s gone. The toner knows.
6	11:33 AM	Choose a Temperature, You Coward!	A Dino Mac & Cheese meltdown in microwave screamo form.
7	11:41 AM	The Cottage Cheese Conspiracy	Stolen dairy, expired regret, and the return of Karen.
8	1:00 PM	Mandatory Misery	HR’s mandatory Zoom about benefits: where souls go to buffer.
9	2:27 PM	Data Wrangled Me First	When your spreadsheet contains both macros and malevolence.
10	4:16 PM	All for One, Nuts for Me	Vending machine injustice set to a bluesy country lament.
11	5:30 PM	Road to Somewhere, Maybe	A hazy drive home filled with existential turns and radio jingles.
12	6:01 PM	I Just Wanted Soup	Too many choices, not enough sanity. A punk breakdown in Aisle 3.
13	6:42 PM	Toast Was Always Plan B	Burned out and hungry, toast is the dinner of quiet surrender.
14	8:48 PM	The Couch Sucks Me In	He sinks. The fan judges. The remote is useless.
15	9:46 PM	Spit in Grief	Brushing with no toothpaste. Cursing the gods. The final decay.
16	2:11 AM	Too Tired to Sleep, Somehow	Insomnia wins. He unravels. Tomorrow threatens to repeat.

The character is pretty much myself – exaggerated quite a bit. In the life of corporate culture, I’ve experienced, seen, and heard of many things. I’ve tried tying them in to the lyrics to help give vivid details to the songs with the goal of making them absurd and hilarious. I would keep looking for more ways to put in personified objects too – while keeping other people out of the story as much as possible. I do have a neighbor that laughs, a “Karen” who disappeared, people in the lunchroom laughing, a manager who sleeps on a zoom call, etc. However, they are more into the background or adversaries so that the individual is “alone” in a corporate world.

A few large language models were used for the lyrics for improvement and feedback. More so as to put in the genre and lyrics and ask “What’s wrong with this?” or “How can this be improved?”. For the most part, it suggested areas with a few options for rhythmic syllables, rhymes, or words that may not be obvious with a few suggestions. I’d pick the best one, or change things around and go through it again. Sometimes I would ask what lines could be removed, as the generated songs were sometimes past the four minute cut-off and just ended abruptly mid-song.

When passing the lyrics through Riffusion, there were problems where the generated singer would skip words, add in extra words, or sing nonsense. In some cases it had quite a bit of trouble as I tried to change words to phonetically sound out what I wanted it to sing. In “Time is a Flat Liar”, I had to write out “I sinked you to the cloud” rather than “I synched you to the cloud”. In some cases, I just couldn’t get it to sing the words correctly. The printer chanting in hex and gold “4B 41 52 45 4E 43 55 52 44 53” translates to ASCII “KARENCURDS”. I could not get the AI to sing it properly. I’ve tried spelling it out, adding dashes, separating each number/letter with spaces, and on and on. So instead we have “4B [cad]1 52 45 4E 43 55 52 [45] 44 53″.

The main problem with Generative music is that you have no control over the singer. You can’t tell it to perform the same song, but change one small part. Each run is a new cover song with different instruments and singers. You can use a “Vibe” or stems to generate similar sounding songs, but it’s not the original. This is the critical part of music generation that really makes it difficult to work with creating music. Had it been a real band playing the song, I could tell them to pronounce or annunciate words differently, swap out words, etc.

Distribution

The main goal everyone seems to have with music is to get their songs to appear on Spotify. You are unable to submit music directly to Spotify. You have to get a distributor to do it on your behalf. Most people who haven’t signed with a major record company do this with distributors that you can interact with directly. DistroKid appears to be one of the most popular distributors. It’s not free, and has an annual fee as well – but you receive 100% of the profits. If you discontinue in the future, you get a percentage of profits.

I had to create a band name, logo, and record label. I created Circuit Dream Records. Since I was often making strange songs, I eventually came up with Bard of Bizarre. For the logo, I used DeepAI – another AI service. Since I am also a clown, I wanted something I could use with a clown performance to put on luggage and business cards if I ever ventured into that area. I’m also a bit of a geek, and have a history with chickens, so I went with the prompt: A coat of arms titled ‘The Bard of Bizarre’ in circus lettering on the shield. The shield features a rubber chicken at the crest. The chicken is wearing eye glasses.

I submitted my album, lyrics, and synchronized the lyrics to the music. Synchronizing lyrics is painful. You have to listen to the whole song and press the space bar to jump to the next line. If you mess up at the end, you have to start over again.

You need to add who played what instrument, and who wrote the lyrics. I put my legal name for the lyrics.

DistroKid then created a HyperFollow page that allowed people to pre-subscribe on Spotify before the album was released. You can also listen to 30 second clips of each song on the album. Afterwards, I created a website, facebook page, youtube channel, tiktok, etc.

HyperFollow: re: Cubicle 13

Website: The Bard of Bizarre

In addition, it showed me all of the platforms that it submitted music to.

Spotify	Apple Music	iTunes	Instagram / Facebook	TikTok
YouTube Music	Amazon Music	Pandora Radio	Deezer	Tidal
iHeartRadio	Claro Música	Saaven	Boomplay	Anghami
NetEase	Tencent	Qobuz	Joox	Kuack Media
Adaptr	Flo	MediaNet

With Amazon Music, I can say “Alexa play bard of bizarre” or “Alexa play cubicle 13” or “Alexa play time is a flat liar”. However, I can’t say “Alexa play [the] bard of bizarre” or “Alexa play [re] cubicle 13″. It’s still pretty cool that I can tell people to talk to their Alexa devices to play my songs.

Over time, I started to see my music appear on other websites

• Lyrics on Demand: bard_of_bizarre
• Musicmatch: The-Bard-of-Bizarre
• Paroles Musique: The Bard of Bizarre
• Lyrics.com: The-Bard-of-Bizarre

Music Video

DistroKid also supports distributing music videos. They are pretty strict on what constitutes as a music video. I’ve seen many people on YouTube who create a static image while music plays. Others try to add things like noise filters that keep changing. Honestly, I don’t like these type music videos when I see them, as their isn’t anything interesting to look at after five seconds in.

Initially, I used DeepAI to create various marionette images to go with the song, “Time is a Flat Liar”. The original video looked more like a slide show. I was a bit intimidated if it would pass the requirements for DistroVid. I then learned how to use various transitions, effects, and filters in CapCut. In the end, I had a pretty good flow that seemed to have more action. The main thing was to take a tip from George Lucas and keep special effects scenes to less than 5 seconds to keep the action moving similar to Star Wars.

The problem that I had with AI images was continuity. I couldn’t have a character in one image appear in the next. I tried going into quite a bit of detail explaining what the character appeared as, but continued to get different results each time. I tried other services, but I couldn’t quite get it to work before my credits ran out. The cost simply to “try” was too high. I also tried animating an image with AI. But that got crazy really fast. In one setting, I explained that the marionette had to wake up screaming as an alarm clock with an evil grin went off. The marionette turned into a person screaming, and the animated scene started to get corrupted.

So in the end, I had many effects with different characters in each scene. I submitted my video and crossed my fingers, and then it successfully went onto various platforms.

Merchandising

In addition to setting up a website, I also created some merchandise people can buy via Zazzle.

After that, I started going back to my roots of board game development, and created a collectable trading card game “Bizarre & Unbalanced” where the goal is to make everyone else have too many points of “Whelm” that they become “Overwhelmed”. The cards are based on the music album, with the ability to easily expand it for additional albums. I got an old Cricut Express working on a macbook and cut out card stock in sizes available on The Game Crafter. I have spreadsheets of card features, titles, descriptions, etc. Many of the card images have been created, including artwork for the card face templates, card backs, directions, and box art.

Burnout

I had a whole schedule of things to do leading up to the release of the album, and what to do afterwards. I was going to create additional music videos, and go through play testing the card game. I did play test it once… However, over time, the library stuff in my local town, the end result of primary elections, the counties loss of the LS&S contract, and one of the county supervisors making claims that the library itself was mirroring conditions of rewriting history based on the book 1984, mainly led up to burnout. It seemed the supervisor reading his statement didn’t quite catch the irony. My comment was praising the double-plus good library. I stepped away from quite a bit of stuff in the time afterward as I recovered from mental exhaustion. The big battle was over, and I needed a break prior to the run up of the November elections and its effect on the libraries future.

Board of Supervisors July 1, 2025 03:28:54 – Cancellation of RFP for Library Services – Jason Ham, County Attorney. Orwellian 1984 references by Supervisor Dr. Richard Jamieson in a 25 minute prepared statement, to argue that a public library operated by a non-profit with a supervisor as one of its trustees is actually a private library among other statements regarding the inability to make FOIA requests between public bodies and financial statements.

My public comment referencing “double speak” and the first mention of the burnt out light and falling ceiling that would continue to be brought up many times in future meetings.