If it sells, it's creative.
Is AI creative? According to the godfather of advertising, perhaps it is. Let's try a little experiment...
I worry that these newsletters may become only negative about AI .. but the reality is we will all have to live with the inevitable change it will bring.
Like all innovation - how we use, regulate and disseminate the innovation affects the outcome, not the innovation itself.
So, for a change, I thought I would share with you one of the ways I’m using standard AI products to conduct a little experiment to answer this question: is AI creative?
I’m going to make a short film - entirely using AI.
David Ogilvy, the godfather of advertising, often said “If it doesn’t sell, it isn’t creative” so I’m going to suggest that if AI can make me an advert, itself, without any creative input from me … and if that advert sells … then I think we can all agree that AI is, in fact, creative.
The project…
In a few weeks I will be launching a new product: 💌 Inbox by Charlie1.
We’re not here to talk about my product but since most of you know that I’ve spent the past five years trying to use AI to unfuck the news … and since I haven’t launched anything to describe Inbox, I’ll leave a little footnote to explain what it is.
Anyway - to launch Inbox, I’m going to build an advert, entirely built with AI … and I’m going to include a few posts along the way to explain how it’s done… because kinda’ the point of the experiment is to show that anyone can do what I’m doing.
We can talk about the morality of the experiment in a later2 post but for now I’m throwing caution – and human creativity – to the wind and letting the machine take over.
The tools, the steps…
I’m taking this seriously. I’m not even writing the script.
I have a rough idea that I need 30 seconds, with a sequence of moving images that tell a story. I know I can generate 3-4 second clips so I need ~10 steps… but aside from that, I’ve left this all to chance – or maths - or AI - or whatever…
The Director: ChatGPT – Writing the script & brief…
The first step is thinking through the advert: the idea, the messaging, the vignettes of graphics and the script for the messaging.Director of Photographer: Midjourney - Casting, costume and scenery…
Using the “prompts” provided by GPT above, I’ll asked Midjourney to generate still images of each of my ten scenes.Cinematographers: AssistiveAI & Runway…
Finally, to bring my scenes to life I’m using Assistive & Runway’s video generators (they have different skills so I’m playing with what works best).
Putting it together…
There are numerous generative AI tools for music so I will come to that at the end – I’ll have to fill you in later..
… and for the final touch [overlaying captions GPT writes and sewing everything together into one video] I’m probably going to dust off my iMovie or Adobe skills. There are better tools, I’m sure…
The rushes…
So far I have to say I’m pretty amazed by what I’m able to do without putting too much thought into it.
To be honest I had started down the road of planning this advert before I decided to fully give myself to the whim of AI - but that gives me something to judge its efforts against. What GPT has come up with, for instance, is very close to my original narrative and if anything I’ve worked to nudge it towards my original expectations.
I will of course share the video when I’m done — too much else going on to be able to focus on it full time — and I’m keeping a bit of a live diary here if you want to follow along…
… but for the time being, please enjoy a little of the “rushes” - the raw unedited footage that I’ve produced so far.
There are plenty of subtleties in how I’m “prompting” (conversing with) each of these tool so if there’s anything you’d like to know more about then please, please just ask in the comments… 💬
Prompt (Written by GPT4) – Generate an image set inside a crowded subway car during rush hour, with diverse passengers of different ages and business attire engrossed in their smartphones and newspapers. Their expressions are a mix of concern, stress, and focus as they absorb the day's news. The interior lighting of the subway car is stark, accentuating the furrows of concern on the passengers' faces and creating an atmosphere of communal anxiety.
[Upload: Image] + A very busy commuter train filled with people reading the news. Dolly shot.
Prompt (Written by Me) – Woman sits in a bustling coffee shop, working on a laptop, surrounded by newspapers, editorial style.
[Upload: Image] + Woman working in a coffee shop, arc shot.
Prompt (Written by Me) – interview room, pleasant setting, a woman being interviewed by a friendly pink robot, editorial style
No prompt… I just gave AssistiveAI the image above…
Finally, here’s a little taste of the conversation with GPT. Unfortunately, because I uploaded some assets for GPT to look at, I can’t share the whole conversation with you (sharing conversations that include images isn’t currently supported for some obscure reason)… so screenshots it is…
Getting started… this is GPT confirming to me that it’s understood the brief so far and asking me a few (pretty sensible) questions for clarification.
We’d done a few drafts and worked through some ideas by this point so here’s the first few frames of our advert properly planned out. You can see the prompt for the first image above in scene 2.
Finally, it’s worth noting that I’m playing around with the newest Midjourney 6 beta for the image generation3. To that end, I had GPT go and read a Medium post with some useful v6 prompting techniques and then flesh out the prompts based on that article.
.. and if that’s something you’d be interested in, drop a comment on this post.. or tweet at me (or X me - is that a thing?)… or something. I want to know what you really think.
I did let GPT4 try generating the images itself with DALL•E but it just doesn’t have the skill or attention to detail that Midjourney has.