How to manage neural networks and solve real issues, not just generate funny stuff

Make a case cover, not thinking about neural networks so far

Each case has Key Visual — a cover, a key visual image that conveys the essence of the case and the main idea of the brand to the reader, and also sets it apart from other projects.

Our first Key Visual looked like this:

We’ve got a joke, we’ve got a logo, but there is no communication on behalf of the brand. Let’s search further. We collected something more fascinating a few hours later.

Now it’s better. Musk led to a rise. Trump led to a decrease. It’s already a story with a composition, and not just a fine picture. But the visual is still not finished.

We’re confused. Hmm, it’s one thing to come up with a funny picture with a story. Another thing is to create it. The designer needs to give more texture. Moreover, funny cases shouldn’t be about Jack and the Beanstalk, but about characters that belong to the market and which almost everybody knows.

Google isn’t an option. It takes a lot of time to sort out the wheat from the chaff here.

Collect facts with the help of GPT

GPT enters the chat, which has helped to get a structured response to an unofficial request.

I ask to provide quotes that caused a radical reaction of the market.

It’s already better, but please provide another approach.

We’ve found a goldsmith. Wow! It turns out that Zuckerberg failed. We do fact-checking and obtain the following.

So, we’ve simplified the search process with the help of GPT. But just placing Zuckerberg on the cover isn’t interesting. We recall some memes about the founder of Facebook and understand: we need a lizard!

Generate Lizards in Midjourney

Being a completely lazy guy, I put the collage above into the neural network and asked to describe it. Hurray, it’s not even necessary to write a text.

We generate it. The lizard, which we will actually use, turned out to be cool on the first try.

All generations further prove that the first option is perfect.

If you demand from a neural network like ‘bring Zuckerberg to me’, it can’t do that. Other tools are needed here.

So, the first lizard is wonderful. I ask the neural network to make the lizard look to the right — and everything turns upside down. The task turned out to be excessive and it’s easier to mirror on your own.

A little retouched — and the KV is ready!

Work with the case body

We continue to tell stories of anti-success and add jokes into boring interfaces.

Zuckerberg and a bubble

The lizard is extremely beautiful (how many times have I repeated this yet?). It needs to be reused. We add it to the case body to enhance communication for those who haven’t understood yet.

‘I don’t think we’re in a bubble,’ Mark Zuckerberg said in 2012 before the IPO, and after going public, the company’s shares fell by 50%.

Moving on.

Let’s ask the neural network to make the lizard blow a bubble — the trailer for this fascinating story you can find below.

Trump and mozzarella

The content of this block was also searched using GPT.

We began with a story about the trade wars that Trump staged during his presidency: he rechecked trade relations with countries and imposed huge tariffs on goods imports.

We ask what impact this has had on the market.

What did Trump say about bitcoin?

The trade disputes that Trump has begun have affected a variety of goods, including Canadian aluminum, but European cheeses didn’t fall under the duty for some time.

We’ve combined the fact with his famous phrase ‘I’m not a fan of Bitcoin’ and voila.

Henry Paulson and a flute

The final story in our case. We’ve also found it in the GPT chest.

We start our conversation with a question: what did important people say before the collapse of the US real estate market in 2008?

Yeah. ‘This is a very manageable situation.’ We request the context of this quote by Henry Paulson, former US Secretary of the Treasury.

Suits, let’s do it.

‘This is a very manageable’ — it sounds like we need it in the context of interfaces with a bunch of settings.

When we generated Paulson with a flute, we learned that no one needs flute players.

Edit with GPT

Neural network GPT turned out to be a good editor. We shorten and rewrite the text in the desired key and accents with its help.

Formulate a communication strategy and think up slogans

‘Write me a slogan’ — it isn’t the way the problem can be solved. To generate it, you need to look at how GPT synthesizes and shortens it all, and then find the very key phrases that we will bring into the slogan construction.

GPT offers ‘Trading made fun’.

Shorten

Fine-tune text

I understand that the title ‘Interface: teach and simplify’ isn’t good. I ask GPT to strengthen headline with information about the benefits.

Again. You can’t say to GPT ‘hey, write me a headline about the interface and make it stronger’ — the neural network can’t cope with this, so I’m giving it a hint. It generates, and I choose a variant that suits me.