Kindly simplify training our own dataset

[NeethanShetty]

Hello, Please try to simplify training dataset of our own. It is taking lot of Time and it is difficult to train using current setup.

[anton-l]

Hi @NeethanShetty, thanks for bringing it up! Now it's possible to use a local image folder to train the model by passing --train_data_dir <path-to-train-directory> instead of --dataset/--dataset_name

Also added a section about using your own data to the README: https://github.com/huggingface/diffusers/tree/main/examples#using-your-own-data

[NeethanShetty]

Hi @NeethanShetty, thanks for bringing it up! Now it's possible to use a local image folder to train the model by passing --train_data_dir <path-to-train-directory> instead of --dataset/--dataset_name

Also added a section about using your own data to the README: https://github.com/huggingface/diffusers/tree/main/examples#using-your-own-data

@anton-l Thanks for your Response. This idea will definitely help.

[NeethanShetty]

But the time taking for training is very large. Please suggest a better option for it

[patrickvonplaten]

Hey @NeethanShetty,

Closing this as documentation to train the model on your own data is now provided. Faster training could include text inversion: https://huggingface.co/docs/diffusers/training/text_inversion