v1.2.1 : GPU option, names fixe, new install.py #34
glucauze
announced in
Announcements
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
GPU
This version adds the ability to use the GPU, a setting that can be configured in SD at startup. At the moment, this feature is only compatible with Windows and Linux, since the necessary dependencies for Mac are not included.
--faceswaplab_gpu options in SD args in webui-user.sh or webui-user.bat or on option is SD settings.
Please be carefull, it will install onnxruntime-gpu that can mess your dependencies. So know what you are doing. See below for SDNEXT
The model stays loaded in VRAM and won't be unloaded after each use. Right now, I don't know a simple way to do this technically, so it's going to take up space. If your system's VRAM is limited, turning on this option might not be the best idea.
I've also made a change that could have some ripple effects. Before, I was automatically tweaking some detection parameters when loading a second model. That's not possible anymore, so I've moved the control for det_size and det_thresh to the global settings. There is auto_det_size option if you don't want to understand that, but it will take more VRAM.
For SD.Next, install dependencies manually :
on windows :
I think on a1111, it should be better.
I cannot enable GPU by default since it might break stuff in SDNEXT/A1111 due to numpy dependencies.
Codeformer
Codeformer is re-enabled by default in global post-processing since it was causing bad results by default.
Names in batch
Names in batch should be kept and prefixed with swapped.
Install.py
I refactored install.py. I hope it will work properly (hard to test comprehensively). Please note that SDNEXT tends not to call install.py if logs did not change, so you may have to install the requirements by hand if you enable GPU.
Beta Was this translation helpful? Give feedback.
All reactions