Gaurav Parmar comments

Results 151 comments of


                                            Gaurav Parmar

keyerror:no clip_txt_model

Ah, that is unexpected, maybe you have some network connection issues?

keyerror:no clip_txt_model

I have not tried using huggingface in offline mode. Maybe this thread might be helpful? https://github.com/huggingface/transformers/issues/11231

Any steps to train with chinese prompt?

I am not sure if this is possible easily as the base models do not support it.

OSError: stabilityai/sd-turbo does not appear to have a file named diffusion_pytorch_model.bin.

It looks like your machine cannot successfully connect to the huggingface repository. Could you try running these two lines to see if you can download the SD-turbo model successfully? ```...

train_pix2pix the result is not good

Hi, I think your training results look worse because of two things: precision="bf16" and batch_size=1. Could you try making the following changes to your setting: -make sure you are using...

paired dataset for pix2pix-turbo

We used the midjourney dataset for the paired training. https://github.com/GaParmar/img2img-turbo/issues/8#issuecomment-2024370487

Can I use the model in Auto1111?

We have not implemented this in this code. Feel free to make a PR if you got around to implementing it! -Gaurav

Trained model for Clear to foggy

Hi @anessabiri, Unfortunately @YanivDorGalron is correct, training unpaired models does require more GPU memory.

Abnormal FID metrics of diffusion-based models

It is because the Instruct-Pix2pix models is trained on a general image editing dataset, and not specifically on the BDD dataset.

night to day

If you are still facing these issues, can you provide some more details about your setup?