Zeqiang Lai

Custom image has been supported

The default GAN model is trained on human face. Maybe switch to model trained on cat could help.

You need to switch to the setup handle points pannel to click points. The black mask is used to identify the editable region, which is not ncessary.

I personally recommend to use Conda to create environment. For you problem, these links might help https://stackoverflow.com/questions/59690698/modulenotfounderror-no-module-named-lzma-when-building-python-using-pyenv-on https://github.com/pytorch/vision/issues/2921 https://stackoverflow.com/questions/72698302/no-module-named-lzma

This is a known issue, https://github.com/Zeqiang-Lai/DragGAN/issues/4 We need some time to support it.

Due to the limitation of GAN inversion, it is possible that your custom images are distorted. Besides, it is also possible the manipulations fail due to the limitation of our...

It actually move a little bit. You need to increase max iterations to make the face fully turned.

See https://github.com/Zeqiang-Lai/DragGAN/issues/14#issuecomment-1559148352

> same shit The latest version would automatically fallback to pytorch native implementation without custom op, which should not need CUDA toolkit anymore. What the error messages do you face...

> On latest ( [4015f1f](https://github.com/Zeqiang-Lai/DragGAN/commit/4015f1fd09857ae788897d105d75669a46da94e4) ) I am now getting this: > > ``` > No CUDA runtime is found, using CUDA_HOME='C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1' > D:\AIModels\Art\GAN\DragGAN\venv\lib\site-packages\torch\utils\cpp_extension.py:359: UserWarning: Error...