Skip to content

Producing AI Cover Songs

It is scary-easy to create AI covers of songs that sound surprisingly good now. I was blown away when I stumbled across songs sung by cartoon voices, celebrity voices, and alternative artists that sounded so good, and wanted to create some of my own. I learned that the process is essentially that you use AI to separate the vocals from a song, use another AI to alter the vocal track, and then marry the two back together again for the final product. I was going down this path when I found Voicify AI, which does all of this easily online, even taking audio from YouTube if you like and leveraging any of thousands of trained voices.

It doesn’t work perfectly every time, but it works very well a lot of the time, and that’s pretty impressive. There are even options to remove background vocals when present to get better results when that is a factor.

I then used MidJorney to create an image to associate with each and then used this free online tool to create a video file using the audio file and image. I did use Audacity to trim empty space at the start or end of a track, but that won’t be necessary if the source audio doesn’t need trimming. I created eight inside a couple of hours and published five of them (the others didn’t sound good). Check them out:

If you want to try this yourself, click here. It does cost $8.99 per month for 35 credits (one per song) which is a pretty good deal given the time you’ll save trying so manually. It is surprisingly easy and fast. I doubt I’ll maintain my subscription; you can cancel at any time, and I’m not sure I will be compelled to create more than the 35 attempts will get me. I think it is a good deal to play with and enjoy deciding what songs will be sung by whom until the novelty wears off. Will it?

Joywave Concert

After repeated attempts to see Joywave with pandemic delays and cancellations, I finally got to see them on April 5, 2023. Easily my favorite band that I’d not seen.

I walked in late and managed to easily get right up close on the side of the stage. As the openers cleared, it turned out the stage was perfectly arranged to my position and my view was virtually unobstructed.

For the first couple of songs there were professional photographers walking in front of me so I wasn’t trying to get a good shot. After the first couple songs, the photographers cleared out and I got a better position where I could hold the phone in a way where it wouldn’t bother anyone. I recorded in Cinematic mode at 1080p.

I’m so happy with how it came out— best show I captured since Spoon.

My Video Created with AI about AI

The potential of AI technology on our world is a topic I am fascinated with, and I’ve been reading and thinking a lot about it it lately. I was considering writing an article over Christmas break which quickly evolved into this: a video to convey some of my thoughts using as much AI as I could quickly throw at it: the images, the phrasing, and even reproduction of my own voice.

I created several images using Mid Journey, prompting for images I thought would be fitting and then used an iPhone app called Werble to add subtle animations to them. I then took my notes and pasted them into scenes using a solution called Descript and had it convert that text to speech leveraging a voice powered by AI which was trained with an hour of recordings of me talking using it’s overdub feature.

To throw a bit more AI at the idea, I took each scene and had ChatGPT rephrase the text of my script. Sometimes I needed to tweak them a bit, but the output is consistently impressive. Below is an example of what I’d get when telling ChatGPT, “rephrase ‘<pasted text>'”.

My Original TextText Rephrased by ChatGPT
When it comes to a service aimed at generating new music featuring specific attributes, the streaming service Pandora has a valuable leg-up on other music streaming services. Pandora understands the makeup of a song with its concept of a music genome project.
When it comes to generating new music with specific attributes, the streaming service Pandora has a significant advantage over other music streaming services due to its understanding of the makeup of a song through its Music Genome Project. This allows Pandora to generate new music that is tailored to specific characteristics and preferences.

I took my notes and had them rephrased by ChatGPT to add an additional role for AI in this effort

The primary idea I propose is that we are likely to see personalized, AI generated content coming to us in the form of streaming services. First music from the likes of Spotify and Pandora, then video from YouTube, Netflix, etc. In a generation or two, this content generated to your specific tastes is likely to become preferred by consumers as much (or more than) that available from human artists. It is both an interesting concept to think about as well as provocative making it a good candidate for this kind of content.

Creating content like this isn’t a passion of mine, so unless by some strange chance it gets considerable positive attention, I don’t anticipate creating more– at least not like this. But it was a fun way to explore the idea of AI generated content, getting hands on with some such tools.

It occurred to me just how easy it would be to plagiarize myself and recreate a new version of this video leveraging AI-based tools. So the next day, I spent half an hour asking ChatGPT to rephrase each paragraph in an authoritative and optimistic tone. I then chose a professional speaker voice (which seems way over the top to me) and published a second version. All new words and a new voice– it’s better in almost every way and took so little effort it is scary…

I actually went and made a third version of this that has a landscape layout and leverages an audiogram template. That’s it this time 😉

NFT Packs

Along the same lines of how its fun to open a blind box and see if you got a good/rare/valuable item or a ugly/common/cheap item (or even a duplicate of a good or bad item), packs work much the same way. In fact, when explaining the focus of my online store, I often use the analogy of a pack of baseball cards which is more familiar to some. I’ve enjoyed opening NFT packs for a while, but only recently decided to give creating some of my own a shot.

First Free Pack

I started with a promotional pack that contained one of two images, a common one with 75% odds, and an uncommon one with 25% odds. They were animated versions of my Craft Beer Spots logo, like so:

I really didn’t market it, but simply posted it on my Twitter and Instagram. And they were gone in a second, so it was clear the bots got them all. But just how robust are these bots? They didn’t just get snatched up, but many were listed as packs, some were opened and the NFTs inside were listed. Lower mints were priced higher, the more rare of the two was priced higher. All things you’d expect, but not things I’d immediately expect from a bot.

But the secondary market activity was really surprising. One pack sold for $15, and the number of sales and prices for a single day was quite unexpected.

The NFTs inside the pack also got some action. The common one has now sold >25 times with a high of $1.50 and the uncommon one has sold 15 times with a high of $3.49. In total, my free pack made over $500 (and cost me about a hundred for to mint on demand and host the drop on Nefty Blocks).

Second Free Pack

My first paid pack was to be Explorer Series 1, so I decided to re-run the experiment with a promotional pack to market this upcoming drop. I created a checklist pack with four rarities that would mimic the design and rarity of the real set: common, uncommon, rare, and ultra rare.

This did even better. It also sold out in just a couple of seconds even though it really shouldn’t have– the 780 packs were not well promoted, but it was just a promo and the simple logo pack had done really well, so it seemed appropriate.

The free pack sold 210 times on drop day which was expected given the previous pack and took in about $620. The pack was where it had success; the cards inside got moderate attention:

  • Common – has sold 23 times for an average of about $0.15
  • Uncommon – has sold 13 times for an average of about $0.10
  • Rare – has sold 2 times for an average of about $0.15
  • Ultra Rare – has sold 2 times for an average of about $1.47

Why is common selling less frequently and for less money than Uncommon? There are more of them, and more in the marketplace. And being cheap makes them a very low risk investment to add to one’s collection. But it does get me thinking– could people have written bots that actually buy and sell between accounts to drive up value and interest? If it is automated and across many projects, it is certainly feasible.

First Paid Pack

I paid Nifty Drops (who took payment in ETH) to help promote this one and also listed the drop on NFT Calendar (but never managed to get “verified” before the actual drop happened so it was of no help). So, with some actual marketing investment and my previous checklist promo out there, I was pretty hopeful there’d be some good activity.

A technical glitch tripped up my launch and so for the first several minutes anyone trying to buy a pack got an error. Talk about frustrating. Despite having plenty of resources for the Drop and for the Mint on Demand Packs, I was getting a CPU error. I couldn’t find anything on the error I was seeing, “Unable to obtain signature from CPU payer”, but I believe the issue was that the account associated with the drop did not have enough CPU. I am not sure because the problem did not clear immediately after I took a few actions I thought might help.

As of the time I’m writing this, I’ve sold 80 packs (click here to see how it is doing now) so far with a majority of the action taking place in the first couple of days. Not near what I had hoped, but not bad. As a brand, this is early days and without a huge marketing budget and my personal rule not to spam anyone, it will take some time to grow.

I had set up a Discord server as this is the primary way I see projects communicate with the community in the world of NFTs. Some of the first feedback I got were users unhappy I wasn’t doing more to discourage bots. Given the exposure of my NFT efforts, I think bot activity has actually proven beneficial but I do feel for someone missing out because they were not there the very second a drop took place for a project like this.

Paid drops don’t seem to be an issue, but if you put up a free drop it sells out in seconds even if you limit 1 per account. The only way to discourage bots from consuming all your free packs is to leverage a white list.

There are two ways to go about this: by requiring the user hold a certain NFT in order to access the drop, or by having their wallet ID on a static white list. I decided to try both…

NFT Prerequisite

First, I tried using my Logo NFTs from my first promo pack as requirements for a pre-minted drop. If you had the common logo, you could get a common NFT, and if you had an uncommon logo, you could get a uncommon NFT.

For the first week, I only had the link visible to those on my Discord server, but that was less than 10 people. A couple came and got theirs, then I made the links public and a few more have picked some up since then.

This method works for sure, but participants have to be well informed to understand the requirements. I’m pretty sure there are no bots taking advantage of this method due to there being such light activity.

Static white list

While a bit more manual, I wanted to give this a try to complete the experiment. I created a #whitelist-request channel on my Discord server for users to drop their wallet addresses. In 2 hours, 130 users had come to add their addresses. I did not even tweet about it, I just created the drop and included a link to the Discord server where they could get white listed.

These must be bots right? This is where I think the evil/unfair advantage of bots may skew. It could be that some have bots with enough intelligence to join the Discord server and get whitelisted, but it is more likely that many people have bots (scripts) simply watching for free drops on Nefty Blocks, so they can take action right away. Simply creating a free drop therefore notifies a sizable number of people right away there is something to come and get; many of which may need to manually take the simple step of whitelisting their accounts.

I realize that Discord has its own mechanisms to help root out bots. A project for another day…

In time, I hope these early efforts are of value to collectors but it has admittedly been a learning process, and one I’m enjoying. I’m planning to take this to other areas in the future commissioning artwork to support projects focused on Craft Beer Artwork and even BlindBoxes. Check out the roadmap here!

Craft Beer Spots NFTs

I’ve been having a good time collecting NFTs on the Worldwide Asset eXchange (WAX). I guess it was only a matter of time before I ended up minting my own NFTs. I pondered which of my content was most appropriate and landed on my Craft Beer Spots effort. I’ve documented hundreds of breweries, bottle shops and bars over the last few years and have taken many photographs of each. The initial concept was that of a “trading card” so I looked through this site and created some collections of related venues to serve as themes for a series. I’ve now executed two series, the first being that of the chosen theme– craft beer spots that specialize in sour beers (among my favorite styles).

Continue Reading →

The Weezer NFT Situation

Like many, I’d been hearing about NFTs for a while and at first it didn’t make sense— and then I was able to wrap my head around it. I’ve always been a collector, so it got my interest. I also play with crypto some so I drank up some rather complicated explanations which I’m sure are a turn off to many.

Continue Reading →

Craft Beer Artwork

I love craft beer and am regularly impressed by the graphics I see on cans and bottles. Some are much better than others of course, but it got me thinking that it would be a nice thing to somehow capture. I had conceived an idea where I might build a contraption that would allow me to move my phone around the cylindrical object smoothly in order to leverage the panoramic photo feature to get a nice flat image.

Continue Reading →

Problems with the new Book Book iPhone case

I’ve been using Book Book for my iPhone case for my last few iPhones now and was cautiously optimistic about Twelve South’s latest design. After using it for a couple of weeks, I reported the following complaints to the company and got not response. I’m going with no case at the moment and carrying a separate wallet; something I’ve not done for several years.

Continue Reading →

Grandson – Live at the UnderGround CLT (Sept 2019)

What a great show. I had the pleasure of seeing Grandson play once a year earlier at the same location when they opened for Nothing But Thieves. I attended that show exclusively to see Grandson, but it is always a bit disappointing to have such a focus on an opening band. Not only does an opening band not get to play long, but the stage setup is is sub-par and the sound is frequently poor as compared to the headliner. I recall at this show in particular, the sound was way too low and so the powerful moments in the songs really lost their impact as a result.

Continue Reading →

The Mike O’Meara Show

I enjoyed the Don & Mike show when I lived in DC, when it was on the air many years ago. I made it to a live show one time, but mostly listened on my drive home from work. A few years back I discovered that Mike had his own show now and I’ve been listening on and off ever since, especially when driving long distances.

Continue Reading →

1 2 4