Borg is Brat: How I created a Charli XCX-Star Trek mash-up with AI
Proof that AI can turn even the dumbest idea into something almost worth watching.
It’s impossible to stay pop-culture relevant in middle-age. Trends move fast, there’s work to do, and if I never understand what “skibidi rizz” means, I think I can live with that.
But I make fleeting attempts to plug into the zeitgeist. I’ve always loved music, so a couple of times a year I binge-listen whatever has caught the internet’s attention. This typically ends with me inflicting an updated version of dad rock on my kids, but this month I found myself down an AI rabbit hole, making a Star Trek parody version of Charli XCX’s song Girl, So Confusing.
In case you make it no further, here is the end result. I present to you Borg, So Confusing.
In this newsletter, I’m going to describe in detail how I took a very stupid idea and beamed it into reality. It’s funny (at least to me), but it also demonstrates:
How simple it is to make a rudimentary deepfake.
How consumer AI tools close the gap between amateur and professional.
How quickly you run into ethical questions.
Bonus - there is a lot of fun multimedia content, including some monstrous AI outtakes.
Borg is Brat
Charli XCX has been making music for 16 years and occupies ground between underground electronica and the edgy end of pop. She makes videos where heads are bloodied, substances abused and, erm, panty mountains scaled.
Charli hit the news in July when she declared “Kamala IS brat” on X, giving the new Democratic nominee a viral moment. Brat is the name of Charli’s new album, and it’s also a newly minted Gen Z adjective (is in “that is brat”, “she is brat”, “brat summer”) that means “someone who misbehaves in a cheeky way and doesn’t conform to expectations”. Like a more fun, less dangerous version of “punk”.
Girl, So Confusing is one of Brat’s many stand-out tracks. It documents Charli’s confused feelings about her relationship with fellow pop star Lorde, anchoring on a single refrain: “Girl, it’s so confusing sometimes to be a girl.” In the version below, Lorde responds to Charli’s honesty with an equally sincere verse about her body issues, and I confess that I had a tear in my eye as they turned a potential beef into a moment of catharsis.
But while I totally accept the song’s premise that it’s confusing to be a girl, there are other experiences in the universe that would be even more confusing. For instance, what if you were assimilated into an alien hive mind but kept your human memories?
That’s exactly what happened to Captain Jean Luc Picard when he was captured by the Borg and forced to become their spokesperson. The line “Borg, it’s so confusing sometimes to be a Borg” started pinging around in the stupider regions of my brain, and a plan began to form.
My objective: turn "Borg, So Confusing" into a full parody song and video over two days, using AI as much as possible. I sketched out a seven-step plan.
Write the lyrics
Train voice model to sound like Picard
Separate Girl, So Confusing’s vocal and instrument stems
Have the Picard voice sing the lyrics to the correct melody and rhythm
Create lip-synced videos where Picard appears to sing
Edit together into a coherent clip
I know this pushes up against copyright and ethical issues. Parody has broad protection under the law in Australia and I’m confident that my use of copyrighted material is covered by fair dealing. But sampling the voice of Picard actor and all-round legend Patrick Stewart is in much grayer territory.
Next week, I’ll write an extended piece about ethics, red lines and potential solutions. But for now, let’s get into the process of making the video.
1. Writing the lyrics (1 hour)
This was easy. I gave ChatGPT, Claude, Llama (via Messenger) and Gemini a document containing the lyrics to Girl, So Confusing, and told them:
“You are a Weird Al Yankovic-style creator who wants to turn these lyrics into a parody. The title is "Borg, So Confusing". It is about the confusion felt after Captain Jean Luc Picard is captured by and turned into a Borg. You must match the number of syllables in each line exactly.
LLMs have a bad reputation as comedians, but I thought Claude and Llama came up with some real zingers. GPT was workmanlike, and Gemini went off-brief (it seemed confused by the syllable command). Here are some representative gags, rated by hilarity level.
I took my favourite lyrics from each LLM and combined them into a coherent song. All models struggled with the instruction to match syllables in each line, so I did need to do some minor editing to get the rhythm right. But all up, this was easy, and the outcome better than expected.
2. Training the voice (3 hours)
Creating a version of Picard’s voice was relatively straightforward, but it took some legwork. I used Eleven Labs, probably the best-known voice cloning software, and gathered some samples of Picard speaking in Star Trek scenes.
Eleven Labs is really easy to use – go to My Voices, click Add a New Voice, and select Instant Voice Cloning. Upload audio files of the voice you want to model, and out pops a voice.
But getting it right took several iterations – my first star fleet captain was trained on samples that contained background music and multiple speakers. It produced a generic American accent.
I needed better training data. So, I attuned my inner Sam Altman and headed to YouTube to pilfer copyrighted material. I didn’t feel very comfortable doing this, but since my goal was explicitly parody, I reasoned the data collection was covered under fair dealing. If a rights holder disliked this, they would most likely send a DMCA copyright claim, and I would just take it down.
I found Wired magazine interviews that featured Stewart talking uninterrupted for minutes at a time, and Patrick Stewart’s Cowboy Classics, an album of country songs made for charity back in 2016, to capture Stewart’s singing voice (which I stripped away from the music - more on that below).
This worked. I retrained the model, and something close to Patrick Stewart’s voice popped out of my laptop speakers.
3. Isolating Charli XCX’s vocals (5 minutes)
This turned out to be ridiculously easy. I used Musicfy, an off-the-shelf product for AI in music, to split the vocal for Girl, So Confusing from the music. Musicfy overdelivered by providing me with stems for vocals, keyboards, bass, drums and more.
I was shocked by how easy this was, but after some Reddit research it seems these tools have been available for six years. They’re getting cheaper and better, and just 10 years ago you’d have needed a skilled professional to even get close. That’s wild.
4. Make Picard voice sing lyrics (4 hours)
I had my Picard voice, but I couldn’t just post the lyrics into Eleven Labs and get what I needed. If you do this, the output departs from the rhythm and sounds a bit like the text-to-speech voice from 90s versions of Windows.
Nor could I use Charli XCX’s isolated vocal track as a model for the Picard voice to copy. I could make the Picard model sing Charli’s lyrics, but the results were nightmarish.
Shudder.
There was no tool I could find that would combine the Picard voice, my parody lyrics and the original vocal track into something that matched the music. But there was a way out. I could sing the lyrics myself.
The trick was staying on the beat - get too out of sync and things sounded messy and flabby. In the end it took around 20 takes to cobble together a track that was strong enough. I’d like to apologise to my wife, my children, and anyone else who listens to the clip below.
But the end result was pretty good! I was ready to move onto the video.
5. Create lip-synced videos (2 hours)
This is another step that turned out to be surprisingly easy. I used the lip sync tool from Gooey.AI. All I needed to do was upload the isolated Picard vocal track, and either an image or a video, and the tool would do the rest.
I don’t really like the lip syncing with images. Everything around the subject remains totally static. It calls attention to the artificiality of the media and feels awkward.
But videos produce more dynamic and interesting results.
I have to admit, my outputs are pretty rough. I was able to improve this by playing around with the padding settings for the mouth, but in my case it didn’t matter too much. I wasn’t trying to pass it off as realistic - I think the roughness adds a comic effect.
I ultimately made five videos of Picard lip-syncing to the parody lyrics, to give myself plenty of choices for the edit.
6. Edit together into a coherent clip (4 hours)
Here’s where things got tough. I don’t have an Adobe subscription, so I used the free Microsoft Clipchamp tool that comes with Windows. It was perfectly adequate, with a simple drop-and-drag interface and more than enough settings and effects for someone at my level.
Clipchamp also has an AI feature that will take your assets and do the editing for you. Of course, I gave it a try, and while I think it will be fine for a simple collage of home videos and stock music, it is not up to creating a complex edit. Here’s the result … I think human video editors are safe for a few years at least.
In the end, I edited it the old-fashioned way. It took a long time to painstakingly match mouth movement to the track and juggle 15-odd assets. But ultimately, I emerged victorious from the bedroom and demanded that my family stop watching TV to admire my masterpiece. (Watch it at the top of the newsletter if you haven’t already.)
Virality hopes dashed
My kids loved the video. They made me play it repeatedly and their infectious enthusiasm made me wonder if maybe, just maybe, I’d created the next viral sensation.
So, I unleashed the Borg on social media – specifically on YouTube shorts and Instagram / Facebook Reels. These platforms have a deal in place with Charli XCX’s record label Atlantic, which meant I wouldn’t trigger an immediate takedown via the platform’s notoriously swift copyright protection algorithms.
The reaction was … OK. My friends were very kind. Some people even seemed impressed. But at present I’ve landed at a grand total of 268 plays on Facebook, 129 plays on Instagram, and 86 plays on YouTube. I tried posting it on some Star Trek meme pages, but there was zero interest.
I guess hope springs eternal.
Up next… ethics!
I had so much fun making this. Kamala Harris called Donald Trump an “unserious man” in her acceptance speech at the Democratic Convention, and I think you can level the same accusation at me.
But I am serious about copyright, free speech, misinformation, AI safety and content moderation. These are some of the most important issues in the world today, and while my parody video is unlikely to be a controversy magnet, it calls into focus a lot of issues that are worth pondering at length.
This includes questions like:
Does Australia’s Fair Dealing exception for parody and satire apply when using copyrighted audio as training data for a synthetic voice (provided that’s what it’s used for)?
Where do we draw the line between free expression (especially satire) and people’s right to control how fundamental aspects of themselves, like their faces and voices, are used in content?
How do we stop people using satire and parody as a fig leaf for distributing misleading videos for political or other purposes?
Let me know if you have any thoughts on this and I’ll try to incorporate them into the next newsletter. In the meantime, live long and prosper.