import { Callout } from "nextra-theme-docs"
It may be a permission problem, please change the model save directory use --model-dir
argument.
- For installer: Rerun
win_config.bat
will install the newest version of IOPaint - For pip:
pip install -U iopaint
The error requires manual modification of the C:\IOPaint-v1\installer\lib\site.py
file (the directory of this file may vary depending on your actual installation path). The modification is around line 173 of the site.py file, changing the following line:
f = io.TextIOWrapper(io.open_code(fullname), encoding="locale")
To:
f = io.TextIOWrapper(io.open_code(fullname), encoding="utf-8")
By default, models will be downloaded to the user folder
- diffusion models:
~/.cache/huggingface/hub
. For Windows users, it'sC:\Users\your_name\.cache\huggingface\hub
- lama and other erase models:
~/.cache/torch/hub/checkpoints
. For Windows users, it'sC:\Users\your_name\.cache\torch\hub\checkpoints
Add command line arg --model-dir=xxxx
Double click win_config.bat
, and you can config all command line args in web UI.
Assume two GPUs, and you want to use second one, add SET CUDA_VISIBLE_DEVICES=1
before IOPaint start in win_start.bat
This is caused by moving the directory after unpacking zip.
Error message when run win_config.bat
:
ValueError: When localhost is not accessible, a shareable link must be created. Please set share=True or check your
proxy settings to allow access to localhost.
Add set no_proxy=localhost, 127.0.0.1, ::1
line to win_config.bat
file
set no_proxy=localhost, 127.0.0.1, ::1
@call iopaint start-web-config --config-file %0\..\installer_config.json
see: Sanster/IOPaint#268
- For windows installer: delete the unzipped folder
- For pip:
pip uninstall iopaint
By default IOPaint download model to ~/.cache/torch/hub/checkpoints/
/ ~/.cache/huggingface/hub
For Windows users, it's C:\Users\your_name\.cache\torch\hub\checkpoints\
or C:\Users\your_name\.cache\huggingface\hub
.
If you used --model-dir=xxxx
to change the model directory, you safely delete the directory you specified.
Some GPUs don't support half-precision floating point numbers (a known issue with 16xx cards),
a green or black screen may appear instead of the generated pictures.
This may be fixed by using the command line arguments --no-half
See Memory