You’re getting this email as a subscriber to the BrXnd Dispatch, a (roughly) bi-weekly email at the intersection of brands and AI. I am planning some corporate AI events as well as my own conference, so if you’re an AI company or consultant interested in connecting with brands, please be in touch.
BLUF: I built a fun AI Manifesto Machine with my friend Michael Grant. You can try it out here.
As I’ve mentioned many times around here, one of the goals for this newsletter (and organization) is to break through the noise a bit by staying as hands-on as possible with the tech. I figure it’s better to get my hands dirty building and talking about that process and what I learn along the way than to pontificate about the latest drama with the OpenAI board.
So, to that end, I present my latest BrXnd experiment: The Manifesto Machine. The idea came from conversations with my friend Michael Grant, a fantastic writer and creative director. At some point, we got to talking about manifestos—something Michael has spent a lot of time working on over the years. And, as tends to happen in conversations like this, eventually, we got to talking about how funny it would be to build a manifesto generator whose main feature was a slider that allowed you to anchor your manifesto to any moment in the universe's history.
And then the idea sat on the shelf for a while as life got in the way. But when I was trying to come up with something fun to build as a kind of holiday present, I remembered our idea and got to work.
The slider came first. It proved to be a hairier challenge than I had initially thought because you can’t just create a linear slider from 13.8 billion years ago to today. Not much happened for those first few billion years, but it would take up the vast majority of the space. So, I had to play with the space a bit more, creating breakpoints in different historical moments. Here’s how it was eventually laid out:
0-5%: 13.8 billion years ago to 541 million years ago.
5-25%: 541 million years ago to 100,000 years ago.
25-40%: 100,000 years ago to 1,000 years ago.
40-50%: 1,000 years ago to the year 0.
50-75%: the year 0 to the year 1400.
75-100%: the year 1400 to the current year (2023).
How do you pull apart a manifesto?
From there, it was onto the good stuff. What are the components of a manifesto? Part of what’s fun and interesting about playing with this technology, particularly building things on top of it, is it forces you to abstract a concept like a manifesto into its parts so that a) you can get them as input from users and b) you can explain them as inputs for the AI.
We started simple:
Brand
Category
Start year (slider)
That worked all right, but the AI (GPT-4 Turbo in this case) was a bit too verbose and wrote in full paragraphs instead of the more staccato style normally associated with brand manifestos. So we added a few more options (for both rhythm and length) and continued to tune the prompt.
But then a new issue arose: the model doesn’t always know enough about the brand, so we needed to feed it some additional information to give it context to work off. To handle that, I wrote a bit of code to grab the top Google results for the brand + category input and return the result titles and descriptions as part of the prompt for the AI. From there, we also added “call to action,” allowing you to specify a specific refrain or CTA, and also a “historical reference” input in case there was a particular event you were hoping to use as your jumping-off point. In the end, here’s how the whole prompt looks:
You are the world\'s best advertising copywriter, and you\'ve been hired to write an anthemic manifesto. Do not make it rhyme or you will be punished. You will be given some criteria to work with:\n- Brand: the name of the brand\n- Product category: the category of product the brand sells\n- Edginess: A 1-5 scale where 1 is incredibly tame, and 5 is the edgiest manifesto the world has ever known.\n- Length: Short = 4 sentences/phrases, Medium = 7 sentences/phrases, Long = 12 sentences/phrases\n- Rhythm: Stacatto = short, impactful sentences, Legato = longer, flowing sentences\n- We Believe: true = structure the middle section with some punchy "we believe" lines, false = ignore "We Believe" instructions\n- Start date: For the start date, choose famous things that happened around that time and use that as a starting point (if the date was 66 million BC, it might something like "Since the days T-Rex roamed the plains," whereas if it was 1492 it might start with "When Columbus crossed the ocean ..." The start date is not when the company started, but rather the point in time at which the story of the manifest should start from.\n- Additional Context from Search: You will also be given some context from Google results about the brand. Use those to help inform the story.\n\nWrite an epic manifesto that would make David Ogilvy and Jay Chiat proud! Do great work, this is very important to me.\n- Additional Context from User: A few optional fields the user can include to help define things like target audience, call to action, or any historical references they specifically want you to hit on in your manifesto.\n\nExtra Instructions:\n-Do not mention the edginess, it is just an input for you.\n- Do not return Markdown.\n- Manifestos should be written with line breaks between each sentence for dramatic effect. This is how real manifestos are written for brands. Do not return double breaks between paragraphs.\n- Do not rhyme. We write serious manifestos.
A few things to note:
It really wants to rhyme. I’m guessing that comes from the way I’m describing length, but I couldn’t find any way to get it to stop without threatening it.
“This is very important to me” was my attempt to take into account some of the research about the kinds of prompts that have been shown to help these tools come up with more reasoned answers (see below).
No matter how many ways I tell it the correct length, it always goes over. I guess it’s like a regular copywriter that way.
Initially, we wanted to have an “edginess” slider but basically found that it delivered the funniest (and best) results when you left edginess at 5.
“Do not rhyme. We write serious manifestos.” Made us laugh.
After the manifesto is generated, a few things happen:
It gets sent back to OpenAI to generate some metadata. Here, I’m using the function calling feature. Specifically, I get back a list of themes, a director who might direct the commercial version, the cinematic style (for the image generation step), and a title for the manifesto.
I send that metadata over to DallE 3 to generate an image. Originally, I was using these as part of the slideshow, but I’m not very good at writing image generation prompts, and I thought it just took away from the whole thing, so now I just use it for the OpenGraph image.
Finally, I send the manifesto text to ElevenLabs to generate a voiceover. I hadn’t played much personally with building things on top of these services, and other than the price (which was a bit crazy when this thing was kicking out 4-minute manifestos), I am pretty impressed by the results. I thought about trying to pick the best voice for the manifesto but eventually landed on using a single voice I liked that I thought sounded “manifesto-y.”
Here’s my favorite manifesto so far. It was before we started to constrain the output, so be prepared for 4-minutes of a duct tape manifesto:
Part of what I love about this is that it’s genuinely fertile creative territory. The idea that duct tape could be the enemy of the entropy born in the Big Bang is fantastic and fun and just weird enough to work (if you’re the CMO of Duck Brand, I’d love to help make this happen).
The idea of this whole thing is very much not to replace copywriters or anyone else but to find the opportunities and boundaries of these tools by playing. To that end, here are a few interesting takeaways for me:
Just like real writers, this thing has an obvious set of crutch words it relies on. It forges things in the crucible of primordial alchemy.
To the entropy/duct tape point, I think some legitimately interesting insights can come out of an approach like this. Ultimately, brand communications is built on some connection between brands and culture/history, and being able to kick out a bunch of those, maybe even relating to random points in time, seems like a fun and interesting way to start the kinds of conversations creative directors, strategists, and CMOs are having daily.
The AI voiceover stuff from ElevenLabs is pretty amazing. I’m impressed that it can generally keep a good cadence and deal with some basic intonation. It’s definitely not a human, but it’s pretty incredible how lifelike it is, particularly compared to the stuff that came out of OpenAI recently.
For as much as I think the whole idea of prompt engineering is dead in the day-to-day use of something like ChatGPT, it’s very much alive when building tools like this. Small changes in the prompt can lead to significant outcomes. We also tried to do some things that we just couldn’t get the prompt to interpret correctly. For instance, we were trying to add a “confrontational” checkbox, but the AI kept coming back by talking about “confrontation” but not representing it in the text. This generally fits an overarching experience I’ve had with these things: you frequently need to approach prompting from an angle to get the best results.
That’s it for now. Go have a play, make some manifestos, and send your favorites my way.
As always, feel free to be in touch if there’s anything you want to talk about or I can help with.
— Noah