I think that the title is quite self explanatory! :)
We’d like to serve SD 2.1 first, then ControlNet model on GCP, so I am wondering if I should do it with a regular fastAPI backend on some compute resource or if there is some nice ML serving component that could do the job more easily?
Thanks in advance!
Great topic! I'm very interested in Stable Diffusion and right now I'm running it on my local PC.
Google Cloud has the A2 and G2 instances family that are pretty capable of running SD, since they have NVIDIA GPUs.
Take a look at this link. https://cloud.google.com/compute/docs/gpus
I also found a few links that could be useful to you
Keep us posted about your findings on running SD on GCP.
Let's see if there are other members who can add something to this discussion.