Hello,
I am looking for a methodology or a scientifically backed-up abacus file / strategy to evaluate the carbon emission of the generation of image and videos per request (per inference) on very closed models where we don’t have data (such as Nano Banana, Imagen, etc.)
I’ve identified Ecologits but it seems based mainly on text-to-text generative models, and does not provide material for image and video.
I only have access to the electrical general consumption of our global SAAS platform, and not much more : do you have any methodology that apply to this type of context ?
Or do you have abacus that can help estimate the size and caracteristics of models (such as Google ones, etc.) ?
Thank you for your help 