AI driven technologies do have an environmental cost though, and my attitude towards sustainability made me want to add another layer of meaning to these photos.
After some research I found another OpenSource python software ‘CodeCarbon’ that estimates the amount of carbon dioxide (CO₂) and power consumption produced by hardware. I ‘monitored’ my RTX 4090 GPU while ComfyUI was running. The first image generation usually takes the longest—around 2 minutes—while the following ones are much quicker, just a few seconds each.
Meaning, the power consumption drops after the initial generation is done. In the end, I used 5 flower images out of a total of 89 generations for my project. After retrieving the kWh information from the software and multiplying it by the recent Netherlands carbon intensity rate [455 g CO recorded on 26/02/25] - the carbon footprint calculations looked something like this:
Each of the next 88 images:
1.04468 g + (88 x 0.522795) = 47.05064g CO₂
ComfyUI Workflow Image by Anya Victoria Macdougall
Estimating approximately 47.05 grams of CO₂ emissions by generating images using ComfyUI, considering the electricity emissions produced in the Netherlands was 455 g CO₂/kWh recorded on 26/02/2025.
To put that into perspective, sending a single standard email (without attachments) emits around 4 grams of CO₂, meaning my process had about the same carbon footprint as sending 12 emails.
Now, while that doesn’t seem like much, when you consider the growing number of AI users worldwide - especially those running models on hardware like mine, things can easily spiral into something way more impactful. Without relying on renewable energy sources or offsetting that power usage, it could genuinely become a bigger issue for the planet.
