invoke-ai
[bug]: If runtime directory exists, configuration script does not create the initfile
Is there an existing issue for this?
- [X] I have searched the existing issues
OS
Linux
GPU
cuda
VRAM
16
What happened?
Until recently, if the ~/.invokeai init file did not exist, the invoke.py script would automatically proceed to create the file (albeit without correct defaults), and continue to load the CLI or web UI.
After a recent change (possibly #1517?), the invoke.py script tries to run configure_invokeai.py to reconfigure the root dir. But if the model config directory already exists (specifically, it checks for the configs/stable-diffusion/v1-inference.yaml file), the logic in the configure script bypasses the code path that creates the initfile altogether.
This means if we're missing the init file, but have an existing runtime dir, the application can never load, because it goes into an infinite loop with emergency_model_reconfigure, looking for the initfile but never creating it.
Potential workaround is to comment out https://github.com/invoke-ai/InvokeAI/blob/development/scripts/configure_invokeai.py#L697, which causes full re-initialization of the runtime dir even if it exists.
Screenshots
No response
Additional context
No response
Contact Details
No response
I started fixing this in #1615, but still not quite working because Args.parse_args() uses different logic for finding the init file. Can aim to finish it later today... :sleeping: