Hacker News Reader

HNHacker News

Top New Best Ask Show Jobs

Comment by lysp | Hacker News Reader

Parent Full thread

lysp·

If you have a reasonable video card, you can run it easily locally using this repo:

https://github.com/AUTOMATIC1111/stable-diffusion-webui/

It is extremely active - author updates it 10-20 times per day.

View on HN

TBH, I bought a 3090 and it broke after a week so I sent it back. It was about $1200, and I'm down about $80 on the web site at this point, so... I have been pondering a cloud solution though, but haven't done the math all the way yet. It's a bit like buying a $15K boat and then just going out on it once a month.

> and it broke after a week

Have you monitored its temperature while using it? Did your warranty cover it?

SD even runs fine, although slower, on MacBook Air M1 thanks to quite capable GPU which shares RAM as VRAM, I think soon to utilize Neural Engine Cores too. Announcement and discussion: https://www.reddit.com/r/StableDiffusion/comments/xbo3y7/one...

Slower means around a minute per picture. It's amazing! Guys with GPUs more expensive than my whole computer can't run it, but a laptop without fans can. Well done Apple.

I have a old ass GTX1070 and run it just fine, it's not the fastest but it works. A lot of these cases are just PEBCAK's and should just be removed from the internet, it's been made so easy that a literal child could install and use SD.

Well as long as they have a recent-ish NVidia card, rip AMD users.

The 3090 I had was actually somewhat slower than dreamstudio.ai, but the nice thing was the set of python tools I got with it was scriptable so I could do things like create an image, use that as input for another one, and so on, then make them into an animation. There's a bit of tax with initializing python and loading the ML model with each iteration, but if I ever get my hands on a 4090 I'm sure I can solve that.

IMO the local UIs are still designed like a slot machine :)

Reply on news.ycombinator.com