At CWP we love telling stories using real people – they are authentic and emotional, and most of all, we believe in the power of communicating real-world solutions around the world’s greatest problems. If we can demonstrate to others how to better support people and the planet, then we can make a positive impact.
But character-driven stories can take a lot of time to make, and they are expensive – both for us as a business and at a cost to the planet. While we have strict rules around air travel and closely monitor our footprint (this website is also regularly checked) it’s hard to avoid travel to tell the stories we are passionate about.
We also know that there are many forms of storytelling. To reach a broad and diverse audience, it’s important to create different stories.
So, this Earth Day we decided to try something a bit different. What if we could use Generative AI to tell a different story? One that exudes optimism (one of our core values) but also lets us test a new kind of creativity using nothing but a brief, one person, and a computer.
We were pleasantly surprised by the result. It’s cinematic, emotional, diverse, and eerily global. Locations feel far and wide. However, we also know that using Generative AI can be controversial – so we wanted to take a deeper look into what we created. While we know this technology has the power to displace jobs (and create new ones), we wanted to understand if there is another side to AI. How green is this technology?
Does the energy it takes to compute each word, image, and sound outweigh other methods? But also, does this new creative form have the ability to help us reach new audiences?
To find out, we broke down the production, researched the digital footprint, and reflected on the creative result…
Producing a Generative AI film – how did we get to the end result?
Firstly, like any story we tell, the project began with a creative brief. We kept it simple — create a film for Earth Day that shares our optimism.
So, we briefed Chat GPT with a simple prompt — “An optimistic monologue for Earth Day about saving our planet from climate crisis”. This was honed with five more prompts, then tweaked with some new words and phrases — a combination of human and Chat GPT support.
Afterwards came the “production” phase, starting with audio. It consisted of recording a real person reading the script to give it the pace and intonation we wanted. Next, it was uploaded into ElevenLabs where we selected a voice we felt hit the right tone. We looked for some AI music options, but ultimately this proved not as easy as we had hoped (although this space is also developing rapidly). So, we decided on a track from our Artlist library instead. We also added sound effects.
Next, the visuals – a much more laborious task. Creating the visuals was a two-part process. First, we had to decide what to prompt to match the voiceover, and then decide how we wanted to animate it.
In Midjourney we created 500+ images that were whittled down to 84 upscaled images. There was some clean-up and tweaking required in Photoshop before we used Runway to see which ones animated most naturally. The 84 images resulted in 124 motion tests as the output would differ depending on the settings. From there, we ended up with 27 final motion shots that were taken into Final Cut Pro to create the edit.
What was our Digital Footprint?
We wanted to compare the AI production to a similar film using real people and destinations that would mix shoot footage and stock videos. In a real production, the assumptions are that we would source a voiceover artist to read our script but utilise the same Artlist music and sound effects.
Comparing the two methods was not easy – a lot of the research and information online around Generative AI energy consumption has wide discrepancies. We found a recent study done by Carnegie Mellon University that MIT Technology Review covered last year. It felt like a safe source.
According to this article – “Generating 1,000 images with a powerful AI model, such as Stable Diffusion XL, is responsible for roughly as much carbon dioxide as driving the equivalent of 4.1 miles in an average gasoline-powered car. In contrast, the least carbon-intensive text generation model they examined was responsible for as much CO2 as driving 0.0006 miles in a similar vehicle.”
In looking at what we created, the script and images are equivalent to driving around 2 miles in a gasoline-powered car. It doesn’t include the motion effects in Runway that are energy-consuming (information about this specific area of AI energy consumption is limited — if someone knows, please share the data). A Midjourney image takes around 45 seconds to create, whereas a motion effect in Runway takes around 3 minutes. So, we felt that we needed to increase our mileage to 50 miles — even if this is somewhat of an unknown, it gives a fairer perspective.
On the flip side, if we were to make this film using people and locations, what would that look like? Being conservative, we could split the film into two halves: UK shot footage and stock video. That’s around 13 images we would have to find locations for and drive to. If we were to assume a 50-mile radius and combine some locations, we would be looking at a minimum of 350 driving miles in that same car. Then, the talent needed to travel, and the filmmaker who shot the stock images would have needed to travel (hopefully not by aeroplane).
So, on paper, AI is much greener.
However, this doesn’t consider the wider picture. While it may have been greener to make this film, there is a much bigger impact AI is having on the planet that is hard to research and calculate. For instance, we have to take the water it takes to cool all of the servers into consideration. Then there is the computing power to train AI in general. Nor have we taken into account the behaviours around AI, as millions of people are now generating images and text across the world – a lot of this creative output never being seen by anyone. Essentially, a creative AI wasteland!
All that said, we remain cautious of this new technology. While we can see the value for projects, we think it should be used strategically and for specific tasks that help improve our storytelling.
The creative output – does AI have the power to engage and reach new audiences for CWP?
We will always be passionate about telling the stories of real people. We believe nothing is more powerful than to see and hear what is possible through someone’s learnings and experiences; demonstrating to other individuals and organisations what they can do to be greener and help avert the negative impact of climate change.
To that effect, the result of this AI film was very different from what we usually create at CWP. While it lacks the authenticity of real people, it is engaging. The film does make us feel emotions. It creates hope and optimism and reminds us that while climate change is impacting us in negative ways, there is still a light at the end of the tunnel if we come together, act wisely, and use technology for good.
Will this film reach a new audience? Time will tell, but we hope so, and welcome people’s comments, and are eager for it to start a discussion and raise awareness for the future of energy usage in the AI space. We’ll be sure to cover this in an additional note below in a couple of weeks!
For now, thanks for reading. Watch the film, share it, and have a Happy Earth Day!
