There's a lot to learn from Coca Cola's Godawful AI Commercial

I think that in watching Coca Cola’s disastrous AI Christmas adverts, I had the same thought as 99% of other people - why?

My take is that when it comes to AI imagery, nearly everyone is turned off by the creepy AI look, and the questionable ethics of it. Not only did Coke make a whole AI advert, they even added text at the start of one of them to make sure that everyone knows it! Maybe it’s some rage-baiting marketing angle, I don’t know, but I’m sure that even they were taken aback by just how unwaveringly negative the response has been.

People hate the idea of a company that size trying to replace humans when they have no financial need to do so. But the fact that they tried, and failed is very positive, surely? What better way to say “AI isn’t replacing you yet”?

I’m not going to try and predict the future, but as I see it right now, trying to make adverts or films using AI is a completely misguided approach, and comes from the same lazy approach as Hollywood using VFX for everything and ballooning their own budgets rather than…you know, planning stuff out and thinking a little bit.

AI (and VFX) makes much more sense to me if it’s used as a tool. And for it to be a useful tool, it needs to be controllable, and it just isn’t. Not really - you can sit and try different prompts all day, hoping that it’ll get you something halfway useful, but that’s not really control, is it?

In VFX, it’s the opposite - you have pin-point control over what you’re doing. If you’ve been joining or watching the Follow Along sessions, you’ll see how in compositing software we can build up a large amount of nodes fairly quickly, and at a glance that can be quite overwhelming. But if you’re doing it correctly, every single one of those nodes is there to control something, and that’s ultimately where your creative freedom comes from.

My forays in AI have been with that same ethos in mind. How can I control this mindless beast? The most logical thing to me is to learn from Coca Cola and do the opposite. Don’t try to use generative AI for everything. Use it to create the smallest things that you can control.

If you didn’t see the newsletter where I used an example of using the AI model Stable Diffusion in Blender, you can see that here. Using it through Blender allows you to use basic geometry as a guide to the composition - to me that’s an absolute game changer, because using visual input to control a visual output makes everything easier! The same goes for Adobe Firefly, which at least lets you choose where you’d like the generated output to go (there’s also a newsletter using that here.)

One thing that I’ve been experimenting with recently is asking AI to create outputs on “flat green backgrounds”, so that instead of having one huge, wobbly, warping shot from hell, I just have a single element, which I can then integrate into my work using basic VFX techniques.

In doing this, I’ve actually gone a couple of steps further in working out how to control those elements, which I’ll break down into a practical example next week. There’s definitely an upside to using AI for that kind of thing, rather than trying to find it in stock footage!

You can work the other way too, and create your base plate in AI, and treat it with VFX. I’m certain that all the text, billboards, and animated lights in Cola’s adverts were VFX on top of AI. My gut tells me that you should only do that if you absolutely can’t get the base shot any other way, because I imagine that removing that AI feel/smell from a shot is actually a tonne of work. Let me know if you’ve tried it, or seen anyone do it successfully!

 I hope you have a great weekend, and find time to sit back, relax, and enjoy a Pepsi…