>>936157318
>>936157004
I'm fairly new to lora training myself, but when it comes to the dataset, I just make sure each image is above 1024 pixels and as high-resolution as possible. It's better having 20 high resolution images over 60 mid ones. For cristinysam, I noticed her twitter images were much higher res than the ones on insta, so I sourced them from there. As for tags or captions, I personally go with tags since I prompt my gens using the tag format