/vt/ - Virtual Youtubers

Personally I just think some have developed a sort of stockholm syndrome with the look of previous generation AI art that was painfully obvious just from its style. Consistency was a strength but boy do I hate how literally every model looks without style loras, with a tolerance specifically for the block merge of based64/66/67 I made myself, that I still always used with artist loras
NaiV3 is definitely a massive step down in fidelity, every post I've seen looks like they were bicubic upscaled x2, I've been dismissive of it since it dropped honestly. It's a step up in composition but so little gain for what was necessary to train it and what was lost
EF+hll's biggest weakness is backgrounds, though like I've said before I don't really see anyone posting amazing backgrounds and I don't care too much for really detailed backgrounds myself. I wonder how much can be made up by a lora. Still no idea what people mean by "quality". The only thing I miss is specifically how my wada and so-bin loras looked with my mix