Hacker News new | past | comments | ask | show | jobs | submit login

What you're asking for isn't entirely possible for local installs. Yes, you can run SD on a cpu, but each image takes minutes at a time vs seconds via gpu.

For example, it's not possible to run SD on my 2 year old 16 intel MacBook Pro. This is because PyTorch doesn't have support for the slightly older AMD gpu on board. There's a newer framework called RocM for AMD cards that allows them to work with recent versions of PyTorch.

Given all that, the requirements to have a Nvidia card is entirely acceptable, and for the most part a technical requirement.




Minutes isn't really that big a deal though, one could give it a list of prompts to round-robin through and come back in the morning to a huge collection of images to explore. It's just a different workflow.

Ironically, cpu support would be faster for me (in terms of throughput, at least) because I have on the order of a thousand zen cores put only a couple CUDA compatible GPUs with enough ram to run SD.


> What you're asking for isn't entirely possible for local installs.

The project I linked to does it, so it's clearly possible. I didn't ask for speed, I only ask to be able to run it at all.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: