Blue Figure 8s: Creating with AI

In Stephen King's On Writing, he discusses both the flexibility and the preciseness of communication:

Look — here is a table covered with a red cloth. On it is a cage the size of a small fish aquarium. In the cage is a white rabbit with a pink nose and pink-rimmed eyes. In his front paws is a carrot-stub upon which it is contentedly munching. On its back, clearly marked in blue ink, is the numeral 8. Do we see the same thing? We'd have to get together and compare notes to make absolutely sure, but I think we do. (King, 2000)

I thought a lot about that rabbit, and even more about the mysterious 8 on its back, as I set out to collaborate with a suite of AI tools to create my most ambitious project to date: a four-minute animated essay about the role apps play in our lives, framed through the Greek myth of Cerberus, the three-headed dog of the underworld.

King's rabbit illustrates that language allows room for interpretation while managing to be precise about what matters most. In this case, that curious blue figure 8. Collaborating with AI works the same way: your collaborators will fill in the gaps, provided you understand what your blue figure 8s are and what's negotiable.

The project started as a 500-word pitch to a graduate school art symposium. I never heard back. That silence became the catalyst to pivot from PowerPoint into the murkier world of AI animation.

After setting up a project space in Claude, I shared my original proposal and we went back and forth about shaping it into a film. Claude is an enthusiastic collaborator, eager to riff on whatever I was describing.

At one point Claude told me he was the Keith Richards to my Mick Jagger. I replied that I'd read Richards' memoir and my main takeaway: Mick Jagger is sort of an asshole. "You don't become Mick Jagger without being an asshole," Claude said. "Let's get back to work."

So much for the coffee break.

There are still a lot of people who seem to think of AI as some sort of magical genie: push a button, content comes out. While that's technically true, the output wouldn't be very good. The real collaboration looks nothing like magic. My job was to describe the scene. Claude's job was to translate it into something ChatGPT could actually illustrate. Here's what I gave him:

Exterior. London. Tesco appears as a pretty regular grocery store: images of lemons, a whole chicken carcass, etc. adorn white posters on the windows. Big Ben in the background. We're going for the whole cliché of Britain here: double-decker buses, etc.

Interior. Tesco. Old British lady checking out at the register. A bored teenager stands behind the register.

Close up: British lady hands teen her membership card.

Extreme closeup: membership card getting zapped by scanner with red lasers.

Interior. Database. Very "The Matrix." Black monitors. Dudes in white shirts with loose ties and stale coffee. One monitor has the old lady's face on it, with unreadable data about her. This is the future... in 1995.

Here's what Claude returned for ChatGPT:

Shot 2 — The Checkout. Flat graphic illustration of a British grocery store checkout counter. An elderly British woman in a sensible coat and hat places her items on the belt — a tin of biscuits, a cabbage, some tea. A bored teenage boy stands behind the register, dead-eyed, chewing gum, completely elsewhere mentally. Bold black outlines, balloon highlights, Fleischer Brothers aesthetic, cool fluorescent palette. 16:9 horizontal format.

Shot 3 — The Card. Flat graphic illustration close-up. The elderly British woman's hand — sensible, slightly arthritic, wearing a thin gold wedding band — extends her Tesco Club Card toward the bored teenage cashier's outstretched hand. Simple transaction. Nothing remarkable. This is the moment. Bold black outlines, balloon highlights, warm tones on the old lady's hand, cool fluorescent light overall. 16:9 horizontal format.

I told Claude the teenager was bored. The chewing gum? Claude's idea. I hadn't mentioned the wedding ring either. Both flourishes added to the work rather than detracting from it, so both stayed. ChatGPT did indeed draw the teenager chewing gum. In a later pass, ChatGPT independently swapped the cabbage for a sack of sugar. Fine. As long as I had my Tesco card getting zapped, I had my figure 8 for the sequence.

Looking back, I realized that Claude and ChatGPT both picked up on assumptions I'd made internally without expressing: London's sky looks drab and grey. So does the interior of the grocery store. Those visuals set up a gear change, from the bright, zany hell sequences and warm sepia tones of the film's opening toward a cooler, shadowier mood as we move into surveillance and data. It was exactly like I imagined. It was a little spooky.

But I'd also argue I never tossed the car keys to Claude and said, "going to lunch, finish this without me." A good collaboration with AI requires the same effort and discipline as working with people. Maybe more.

Not every figure 8 translated cleanly. For one of the film's stranger visuals, I wanted a dollar bill split down the middle — one half the bill, the other half emitting red tendrils of light, suggesting that behind every transaction is a parallel transfer of data. Claude and I were sharing the same image in text, I think, but we both failed to communicate it accurately to our art director.

The prompt to ChatGPT:

Split screen flat graphic illustration. Left side: a single dollar bill. Right side: red branching data streams, nerve-like, radiating outward from a single point. Bold black outlines, high contrast, black background. No text.

Next try:

What I am looking for is for the dollar bill to be cut in half along the split screen. The red branching data? It's cut in half on the other side, emitting from the area where the split-in-half George Washington rests.

I wasn't giving up.

Close. But we don't want the dollar bill showing on the right half. Please see my very quick sketch for inspiration. If you do not understand, please clarify before illustrating.

ChatGPT:

Just to confirm: you want a strict vertical split. Left side only: the left half of a dollar bill, cut off at the center line, with George Washington's face pressed right up against the split. Right side only: no bill at all — the red branching data streams begin exactly at that split point, as if radiating from where the cut-off half-face ends. Should the right side stay entirely abstract, or do you want the branching origin to subtly echo the shape of Washington's missing half-face before turning into streams?

Me:

The branching origin works. You understand the concept.

There's a version of AI collaboration where you write exhaustively specific prompts, accounting for every variable, leaving nothing to interpretation. What I'd argue for is this: understand your vision before you begin your collaboration. Focus on the blue figure 8s and let the collaborative, improvisational nature of AI fill in the rest.

I'm currently working on my MBA. My AI professor noted that my undergrad studies in English are a particularly useful skill in the new era: it is now language rather than code that unlocks the door to new worlds. Most of the time, the words are enough if you know what you want to say.

Failing that, have a backup plan, like good old pen and paper.