Use CPU if no CUDA device is detected

[philippschw]

No description provided.

[JustinGuese]

@philippschw how did the duration for the run change? i feel like it would take way longer, if it works at all?

[philippschw]

@JustinGuese, it works but you are also right, it takes over an hour to make a single inference call on my MacBook Pro. 2.9 GHz 6-Core Intel i9 and 16GB RAM.

[illeatmyhat]

That can't be right. I have an i7-1260p with 8 actual cores. A 512x512 image with 50 steps will inference in 12 minutes on my branch.
The OpenVINO model does the same in 7 minutes.

[JustinGuese]

yeah anyways, cpu time is way too long

out of curiosity: if I would provide a "hosted" stable-diffusion, would you actually pay for it? or rather use the CPU for cost reasons? like the server would cost me ~600€/month, so I would need a lot of interest/people to be worth it

[shi-yan]

conda env create -f environment.yaml

fails on mac, not sure if this PR should include an alternative environment.yaml?

[shi-yan]

conda env create -f environment.yaml

on mac, not sure if this PR should include an alternative environment.yaml?

sorry, didn't realize there is environment-cpu.yaml

[jpdeleon]

There is a bug: undefined variable wm_encoder in scripts/txt2img.py, because imwatermark was intentionally removed.

[philippschw]

There is a bug: undefined variable wm_encoder in scripts/txt2img.py, because imwatermark was intentionally removed.

Thanks @jpdeleon, I updated the code.

[EonDigital]

This is a non-additive interface change. I'd recommend against as it impacts anyone writing scripts to interact with the original code. It's also beyond the intent of the commit.

[EonDigital]

Where is your .gitmodules file?

[alperinugur]

That can't be right. I have an i7-1260p with 8 actual cores. A 512x512 image with 50 steps will inference in 12 minutes on my branch. The OpenVINO model does the same in 7 minutes.

similar for me here