Just noting that image gen is done by s different architecture than an LLM, and the environmental impacts are not the same (in particular, much much less energy use than having a human make one image, and substantially less water use than going to the toilet).