Thanks for your question.

It mostly means ethnically diverse. This is what the dataset focused on. Both for characters and the environment.

There are a few images of different age, and very few of different body weights/shapes/sizes.

I'll upload the dataset for it so you can check it out.

@Janet I think that's a very good point and I hope that sharing mine will help a little bit.

Regarding body types and ages that's a good thing to try to balance for the future.

With this model, I mostly focused on picking as many varied images as possible, and the dataset is all generated using a prompt containing a random nationality of the entire world.

But what I don't have is a large balancing system to make sure there's an even divide, and especially across body types. You would have to work extra hard to strike a strong balance in that, and maybe you'd even need to overcompensate, to even out the originally traind biases.

I haven't collated and uploaded it yet, but I'll try to remember to ping you here when I do.

I will update my Dataset Article (https://civitai.com/articles/2138/lora-datasets-training-data-list-civitai-dataset-guide) when I do, so if you subscribe to that or however that works, you should get a notice whenever I update it I think.

@Janet Correct! All generated dataset. I don't have the rights to use thousands of professionally taken photos :)

Although there are probably high quality royalty free stock photos as well. But with generation, I at least know where one layer of the process comes from.

The question is what the models I use are trained on... and what the models they used trained on, and the base model, etc.