A technology I find problematic — and use every day anyway

AI image generation consumes energy on a scale I'm not going to downplay. Data centres are among the most resource-intensive infrastructures we've built, and the appetite grows with every new model generation. I haven't owned a car in years, I travel by bike and train, I don't fly, and I try to consume as little as possible. And then I sit here writing about how to integrate AI models into professional image production. That contradiction doesn't resolve. I've tried.

Local inference instead of cloud APIs makes a real difference — every render stays on my own hardware, no data transfer, no remote data centre. Using small, already-trained models rather than constantly fuelling demand for the next biggest thing means: no further training is driven by my demand. A targeted workflow instead of a hundred blind render passes, compute-heavy steps only for the final result — one or two passes instead of twenty — reduces the actual energy consumption per image considerably. These aren't symbolic gestures. They are decisions that make a tangible difference in everyday work. But they don't solve the underlying problem. The training of these models has happened. The infrastructure exists. I'm part of this system.

The second aspect matters just as much to me, maybe more: AI is accelerating the concentration of economic power in the hands of a very small number of corporations in a way I consider dangerous. Whoever controls the models increasingly controls the visual language, the tools, the dependencies. That's not an abstract problem — it's actively changing the conditions under which professional image production happens.

That's why choosing Blender, ComfyUI, SDXL, and Flux isn't purely a technical decision. Open source is a position here, not a licence model. Tools that work without subscriptions, without corporate entanglement, without lock-in — that's an attempt to at least minimise one's own dependency within a problematic system.

And then there's an argument I have to be honest with myself about: if I don't teach this method, someone else will — someone who doesn't ask these questions, who draws no consequences, who simply recommends the next big model because it produces impressive images. That's not a justification. It's the reason I continue anyway, even though from my perspective the line has already been crossed.

I don't have a solution. What I have is a way of working that tries to minimise the damage — and the conviction that engaging consciously with a problematic technology is better than looking away.