This is truly a delightful job. Since I can't use the 'pipeline' method to generate images, I wonder if I need to prepare some pre-trained files on my own computer, especially the environment and configuration for the 'Kandinsky model'. Or can I train it from scratch on my own without the support of other resources?
Training new operators would require either the pretrained models which we shared here or training from scratch which would require GPU resources and data
This is truly a delightful job. Since I can't use the 'pipeline' method to generate images, I wonder if I need to prepare some pre-trained files on my own computer, especially the environment and configuration for the 'Kandinsky model'. Or can I train it from scratch on my own without the support of other resources?