A human drawing a thumbnail in 15 minutes consumes 0.025 kWh. An AI creating an image consumes between 0.06 and 0.3 kWh, so between 3 and 12 times as much. Both have massive supply chains that go into producing and maintaining them.
I don’t really have a horse in this race eitherway, but what about finding a person who can draw a decent looking thumbnail in 15 minutes? Probably that’s gonna be using various webservices such as fiverr or something along the way?
But the whole idea of comparing them is kinda funny. As if that human would just be turned off and not consuming any energy if they weren’t making a thumbnail for your blog. Though maybe they’ll make a cup of coffee they wouldn’t have otherwise before getting to work. You never know!
If you’re going to factor in the energy already expended whether or not db0 generates an image to calculate the energy of that image, then you’ll have to amortize it across whatever millions or billions of images generated using that model.
You certainly have to account for it, but surely some consideration must be paid to the fact that most of these images are quite useless. I’ve probably generated 100s of midjounrey pictures myself, to very little benefit except seeing what it was capable of. If we treat these usages as equal to cover image for a blog, I don’t think it’s quite fair. Not to mention the actively harmful usages (CP, deep fakes etc).
I’d say the generated image for the blog and any other drawn or photographed image for the blog are equally worthy or worthless. You’re there to read the article, the imagery is entirely superfluous if it can be adequately replaced by a generated image.
If you’re talking about needing to generate 5 images to get one to use for the article then yes you’d sum their adjusted costs but those images you generated for funsies are still accounted for because they served their purpose.
15 minutes is crazy fast and assumes they just get exactly what they want first go. You need to factor in running your PC using Photoshop or equivalent, which is fairly resource intensive, sustained over what is realistically 40-60 minutes at best, sourcing assets from servers which are using energy to serve the images. Compared with AI which has high usage for sure but it’s extremely short bursts.
A human drawing a thumbnail in 15 minutes consumes 0.025 kWh. An AI creating an image consumes between 0.06 and 0.3 kWh, so between 3 and 12 times as much. Both have massive supply chains that go into producing and maintaining them.
I don’t really have a horse in this race eitherway, but what about finding a person who can draw a decent looking thumbnail in 15 minutes? Probably that’s gonna be using various webservices such as fiverr or something along the way?
But the whole idea of comparing them is kinda funny. As if that human would just be turned off and not consuming any energy if they weren’t making a thumbnail for your blog. Though maybe they’ll make a cup of coffee they wouldn’t have otherwise before getting to work. You never know!
Everyone is a NPC, they only activate when you’re close to them or when you contact them, otherwise they are in standby mode
Not to mention the power requirements for transferring funds digitally
I can make a GenAI image on my PC in 3 seconds. 0.06 kWH is outrageously wrong.
including the training data?
If you’re going to factor in the energy already expended whether or not db0 generates an image to calculate the energy of that image, then you’ll have to amortize it across whatever millions or billions of images generated using that model.
You certainly have to account for it, but surely some consideration must be paid to the fact that most of these images are quite useless. I’ve probably generated 100s of midjounrey pictures myself, to very little benefit except seeing what it was capable of. If we treat these usages as equal to cover image for a blog, I don’t think it’s quite fair. Not to mention the actively harmful usages (CP, deep fakes etc).
I’d say the generated image for the blog and any other drawn or photographed image for the blog are equally worthy or worthless. You’re there to read the article, the imagery is entirely superfluous if it can be adequately replaced by a generated image.
If you’re talking about needing to generate 5 images to get one to use for the article then yes you’d sum their adjusted costs but those images you generated for funsies are still accounted for because they served their purpose.
15 minutes is crazy fast and assumes they just get exactly what they want first go. You need to factor in running your PC using Photoshop or equivalent, which is fairly resource intensive, sustained over what is realistically 40-60 minutes at best, sourcing assets from servers which are using energy to serve the images. Compared with AI which has high usage for sure but it’s extremely short bursts.