Are we already paying for "Free" AI photography? The hidden costs of generative AI laid bare

We’ve kind of got used to the idea that AI in photography is free, as more and more AI features are added to regular photo-editing software. But maybe this is just the first phase. Maybe this is a long-game business plan we’ve seen before, where we’re all drawn into something that’s ‘free’ until companies unveil their plans for monetising it and it’s too late for us to get out again.

So how long will it be before we have to start paying for Generative AI imaging technologies, how much will it cost and will it be worth it?

How much will you have to pay for generative AI?

Building and running AI data centers is expensive in all sorts of ways (more on this shortly) so software publishers have been quick to find ways to fund it. Topaz Labs is an AI trendsetter, now offering Cloud Credits for faster cloud-based AI processing in its Photo AI software, and pay-to-use AI web apps based on subscription with commercial usage and resolution tiers.

And while we’ve got used to generative AI tools in Photoshop, Adobe’s online Firefly 4 model needs a paid Premium subscription for generative video. Will that filter down into the photography sphere one day? Well, the bad news is it already has. The main image for this article apparently cost me 1 credit to create. You get 100 credits per month with the Photography Plan and Lightroom plan. Adobe is turning Firefly into a separate product.

ON1 Software, meanwhile, prides itself on its subscription-free approach but builds in an optional third-party Stability AI subscription for those who want enhanced Generative AI Crop and Generative Erase results.

Don’t be surprised if ‘free’ not terribly good generative AI is used as a tempter for a paid, ‘good’ AI upgrade. That ‘freemium’ pitch ain’t new.

How much will your conscience pay?

Let’s not get too bogged down in wishy-washy philosophical debate because everyone has their own moral compass and probably isn’t too interested in other people’s. But I can say from personal experience that the more I manipulate an image, the less personal and emotional feeling I have towards it. If that kind of thing matters to you, then generative AI is perhaps just one step too far.

If you want popularity and likes and influencer income on social media then generative AI is a godsend; if you value your personal connection to the physical reality you inhabit, then not so much.

How much will the planet pay?

Generative AI, like AI processing in general, takes a lot of processing power. So, like, whatever… who cares? Tech gets better all the time, computers get faster, nothing bad has happened, it’s a big tech problem not mine, right?

Well, hold on. Current estimates suggest that an AI processing search query takes about ten times the electricity of a regular query, and that it’s just the energy requirements we need to think about but the land, water and materials needed for the construction of semiconductor plants and data centers. This is according to a June 2024 analysis from Barclays Research, which suggests annual US data center power demands could grow from 14%-21% every year to 2030. US data-centre demand could triple by 2030, from 150-175 terawatt hours (TWh) in 2023 to as much as 560 TWh – equivalent to 13% of the current US electricity demand.

According to a BBC News analysis, the AI industry could be using as much energy as the whole of the Netherlands by 2027. The article also quotes Danny Quinn from Scottish data centre firm DataVita, who said "A standard rack full of normal kit is about 4 kilowatts (kW) of power, which is equivalent to a family house. Whereas an AI kit rack would be about 20 times that, so about 80kW of power. And you could have hundreds, if not thousands, of these within a single data centre."

According to this MIT Technology Review, Meta and Microsoft are looking at new nuclear power plants, while President Trump’s Stargate initiative has set aside $500 billion for up to 10 data centers each requiring as much as 5 gigawatts of power – more, allegedly, than the total power consumption of New Hampshire.

Let’s not get political. This whole area is going to be surrounded by controversy, claim and counterclaim. The truth will emerge eventually. But what is clear is that AI (including the kind of AI we now use in photography) is not just a purely digital tech with no physical impact on the world.

Something to think about, right?