Arch Linux User Repository

Note: on M1/Apple Silicon via Asahi Linux, performance is very slow. ~28seconds per iteration. armpl libraries are installed.

The current pkgver() implementation is incorrect - it generates multiple versions each on a newline, causing the build to fail (whitespaces aren't allowed).

How do I specify another location for the "Models" folder? Thanks.

Easy Diffusion (stable-diffusion-ui) currently recognizes only my RTX mobile GPU, leaving out my thunderbolt-attached eGPU, integrated Intel GPU, and Movidius (Myriad) Compute Stick 2.

The software doesn't support processing via multiple compute types simultaneously on one device (HSA). For example: although my CPU is visible as an available processor, it's separate from the RTX card and cannot run simultaneous workloads with it.

I previously opened an issue on the opened an issue on the Arch4Edu Github and on the Easy Diffusion Github to discuss creating comprehensive packages like "python-pytorch-all" and "python-tensorflow-all." These packages would consolidate support for different devices, including AMD cards (ROCm), Intel GPUs and Movidius cores (OneAPI and OpenVINO), CUDA cores, tensor cores (TensorRT), and ray-tracing cores (Optix).

A "super tensorflow" package would simplify the experience for front-end users and developers by handling device exposure and references to Torch and TensorFlow.

Arch4Edu suggested that a "fabric" or "virtual backend" might be necessary for multiple Anaconda, Docker, or mixed virtual environments. Each environment would utilize different on-device hardware architectures, and the backend would distribute work across devices.

However, this backend doesn't seem to exist at the moment, limiting the current options to either a CPU or an NVIDIA GPU individually processing projects.

Hello "Magnus-Tesshu", by default stable-diffusion-ui uses my AMD CPU. I would like to use my Nvidia GPU instead. How do I do this? Thank you, David.

Remove the package with pacman. Then run sudo rm /opt/stable-diffusion-ui.

That is the only place it puts files. (I'm on my phone though though I might have got the directory name wrong. It is in /opp)

How to completely uninstall everything? I ran the ...server command and it installed some GB's of things automatically. If I remove this package with -Rns, does it remove all such things? I think it probably doesn't,so I am asking for a way to remove everything.

I have updated it. It seems that you have to delete the old files to make it work and have it re-do the setup process. This might be a sign that this package should go out of date? I might do more research next time it updates. I added a small bit to the .install file.

I know that the pkgver and install process is currently a little busted. I would have written a '-git' package, except that upstream explicitely says not to clone the repo in order to use it.

I do want to install the latest version when available as I will not be available to update this often, I will certainly not check as often as upstream releases.

Ahoj,

please add checksum verification since you download a release source code archive. (I suggest to download a fixed version and increase the version each time there is a new release.)

Or convert it to a -git package:

• use git+https://github.com/cmdr2/stable-diffusion-ui.git as source,
• then add git to the makedepends array,
• write a proper pkgver() function that guarantees an increase of $pkgver at every upstream commit,
• rename your package to stable-diffusion-ui-git,
• add stable-diffusion-ui since you seem to download the latest snapshot of some git code: Can you pleaseto the conflicts array.

Thanks for maintaining,
regards!