AI's energy footprint

Honest talk about AI's energy footprint.

Lately I've seen the same comment a lot: "AI art is killing the planet." I live in tropical Ecuador, where every extra degree is felt in real time, so I went looking for actual numbers instead of just feeling guilty. Here's what I found, said gently and without the panic.

Five calm truths I now carry with me:

Training the giant models did use a lot of power, but your daily generations are tiny in comparison. A single upscale or img2img is roughly the same energy as streaming a movie for a few minutes. The heavy lifting already happened long before we pressed generate.

The real energy giants are still crypto mining, endless video streaming, and servers running ads all day. One movie trailer campaign uses more electricity than thousands of us making dragons for a whole month. Perspective keeps the guilt in check.

You can easily cut your own footprint 70 to 90 percent without sacrificing the magic. Lower quality settings when experimenting. Generate smaller and upscale later. Run local tools (ComfyUI, Fooocus, Invoke) when possible. Skip high-variation mode unless you really need it.

The greenest choice we have as artists? Make pieces that stick. One image someone saves as their wallpaper for a year beats ten thousand forgettable ones that vanish in half a second. Depth beats volume every time.

Guilt is a terrible muse. I'd rather plant a tree (I actually did, there's a little cacao sapling outside my window) and get back to telling stories that make someone feel less alone. Creating from joy helps the world more than creating from shame ever could.

So generate your apocalyptic sunsets and praying androids without the weight. Optimize when it's simple, offset when it feels right, and spend the rest of your energy making worlds worth living in.