Suteev, soviet artist style lora, different versions gives somehow different results, v1 and v3 almost the same, imo their only difference is v3 making multiple characters looks better without additional extensions, while still lose some coherence with the style, only minor change was the training scheduler.v1 and v3 making great copy of the style, while being less obedient to the prompt, v6 is vice versa, with different dataset.I highly suggest to add something like this: (multiple girls, 2girls, 3girls:1.3) in your negative for v1 and v3 if you experience too much characters on the generations. Also if you trying to make some fox/cat/etc girls you should consider adding something like (animal, fox/cat/etc:1.3) to the negative, because of the dataset full of the pictures with animals and multiple persons, v6 shouldn't have such problems, dataset for this version was recreated.Trained on NAI checkpoint with NAI vae, so another recommendation to use NAI vae with this lora.

Suteev style LoRA

suteev_v3-000015.safetensors

v3-e15

v1-e8

Files with SHA256: 1fd359d67597eb25352acc6719b0bdaec16aced0eb05fa1222cbef28e04f12e1