Conversation
Fahim Farook
fahim
I have been doing most of my #AIArt generation on macOS using python scripts that I created myself to use the various #StableDiffusion models available out there. However, recently it has become a bit of a struggle to keep up with updating the scripts, generate art, work on my other coding projects etc. (and all this in my spare time ...)
So I figured over the weekend that I'd switch over to using a webUI for StableDiffusion work and that I'd give up on maintaining my own scripts. I decided to go with automatic1111 since that is the most popular webUI around at the moment. Big mistake ...
While automatic1111 is feature-rich and heavily developed, it really doesn't seem to be that well suited for using on a mac. Sure, there are install instructions for macOS and they are supposed to work. But my experience wasn't that great.
First, I ran into an issue on Apple Silicon macs with cryptodome which wasn't even a package directly needed by the webUI but was an ancillary supporting package. No matter what I tried, it seemed to install the non-Apple Silicon version of the package. No info on the automatic site or help about this.
I finally discover that you can't use pip to install cryptodome since that apparently installs the wrong version on Apple Silicon. The only way to do this correctly is to use conda ...
Great! I'd just nuked my conda install since the automatic1111 webUi uses a different virtual environment. And that installation method used pip 😕 So, I wade into the automatic1111 installer code, figure out what is going on, disable the virtual environment creation and the setup and instead, install conda, set up an environment manually and install all the requirements for the webUI. Launch the webUI and it finally works ....
But, try to generate an image using Stable Diffusion 2.1 and it crashes! Hard!! Every single time!!!
I went through this several times over the day and it was a mess. Tried on a Windows machine and while the install ran fine, trying to generate on Stable Diffusion 2.1 again resulted in errors.
Switched to Linux and finally it all worked fine within like 10 minutes. I still got errors and crashes while generating images but the the image generation at least would go through.
I went back to macOS yesterday and got things working somewhat but only with SD 1.5 models. Nothing I could do seemed to make automatic1111 work with newer Stable Diffusion models 😕
On a whim, I decided to try the InvokeAI webui (https://github.com/invoke-ai/InvokeAI) today. Set up a new conda environment, followed the directions to install the requirements, and had the webUI running in about 20 minutes! It's a much nicer UI and it works with SD 2.1 immediately! No fussing about and no crashes when generating images.
So yeah, I'm sticking with InvokeAI for the time being. I'm sure that automatic1111 is technically better and is developed at a much faster pace, but I just don't want to deal with all the issues. If I did, I'd be working on my own scripts instead 😛
So I figured over the weekend that I'd switch over to using a webUI for StableDiffusion work and that I'd give up on maintaining my own scripts. I decided to go with automatic1111 since that is the most popular webUI around at the moment. Big mistake ...
While automatic1111 is feature-rich and heavily developed, it really doesn't seem to be that well suited for using on a mac. Sure, there are install instructions for macOS and they are supposed to work. But my experience wasn't that great.
First, I ran into an issue on Apple Silicon macs with cryptodome which wasn't even a package directly needed by the webUI but was an ancillary supporting package. No matter what I tried, it seemed to install the non-Apple Silicon version of the package. No info on the automatic site or help about this.
I finally discover that you can't use pip to install cryptodome since that apparently installs the wrong version on Apple Silicon. The only way to do this correctly is to use conda ...
Great! I'd just nuked my conda install since the automatic1111 webUi uses a different virtual environment. And that installation method used pip 😕 So, I wade into the automatic1111 installer code, figure out what is going on, disable the virtual environment creation and the setup and instead, install conda, set up an environment manually and install all the requirements for the webUI. Launch the webUI and it finally works ....
But, try to generate an image using Stable Diffusion 2.1 and it crashes! Hard!! Every single time!!!
I went through this several times over the day and it was a mess. Tried on a Windows machine and while the install ran fine, trying to generate on Stable Diffusion 2.1 again resulted in errors.
Switched to Linux and finally it all worked fine within like 10 minutes. I still got errors and crashes while generating images but the the image generation at least would go through.
I went back to macOS yesterday and got things working somewhat but only with SD 1.5 models. Nothing I could do seemed to make automatic1111 work with newer Stable Diffusion models 😕
On a whim, I decided to try the InvokeAI webui (https://github.com/invoke-ai/InvokeAI) today. Set up a new conda environment, followed the directions to install the requirements, and had the webUI running in about 20 minutes! It's a much nicer UI and it works with SD 2.1 immediately! No fussing about and no crashes when generating images.
So yeah, I'm sticking with InvokeAI for the time being. I'm sure that automatic1111 is technically better and is developed at a much faster pace, but I just don't want to deal with all the issues. If I did, I'd be working on my own scripts instead 😛
0
1
5