Kohya S.

Results 370 comments of Kohya S.

I think warning about TensorRT might be ignored, because TensorRT will not be used in training. For multiple GPUs, unfortunately I didn't test on multi GPUs environment. I think Accelerate...

This seem to be caused by the breaking change of `transformers`. I finally found how to fix this and updated `dev` branch of sd-scripts. It will be merged into main...

Could you please share the trained model and the command line to train the model? I will make an investigation.

Hi, Please specify the parent folder of the folder with images (like ``10_xxx yyy``), with ``--train_data_dir`` argument.

xformers officially released 0.0.17, 0.0.18 and 0.0.19. Unfortunately I cannot test with them, but one of them may solve the issue.

Hi! This method is extremely simple. The distortion of composition at large resolutions can be attributed to the U-Net's limited capacity to handle feature maps of such sizes effectively. In...

Hi, thank you for great work! > or we can directly store > > SD15.control_model.weights - SD15.model.diffusion_model.weights > > and any time user load the model, it add the base...

Thank you for merging! As lllyasviel mentioned above, non-transferred models seem to have better or almost same performance sometimes. In my test, openpose seems to generate almost same image. But...

I've uploaded pre-made difference files. https://huggingface.co/kohya-ss/ControlNet-diff-modules/tree/main I have checked each file by generating an image with the extension, but would be happy to check just to be sure.