They are the same datasets in separate folders, like 10_boneyard, 10_tentacles, etc.

@konyconi Thanks for the answer, makes sense cause you're doing the same to untouch the bleeding and a tthe same time unbleed the main concepts. Bleed sub.concepts and unbleed the main ones. Good thinking and yes, it's how I did the Lycoris tests too but with wearables in my case, not styles (yet).

Ah. I forgot to answer your question earlier. Apologies. The tutorials are rather about data preparation (collecting pictures, captioning, etc) than training itself. For the training, I've provided a config file. I'm not sure whether it can be directly used, but you can definitely read the parameter values in the config and set them in the colab. I believe it must work.

@konyconi thx

@memesis I think they work but maybe there are things that you can't find there such as flip augmentation, color augmentation, etc. (Didnt test it on Colab so I'm not entirely sure but some of those didn't appear a month ago cause a friend told me that, lol). In that case you will have things like Cache latents On/off, etc. But most of the config should work. In fact there is no extreme difference having those marked with BF 16 or just normal with FP16. At least with my testings...

@LDWorksDavid thx