If you use my model (<a target="_blank" rel="ugc" href="https://civarchive.com/models/2866668/exercitus">https://civarchive.com/models/2866668/exercitus</a>), in some cases, the LORA strength should be reduced to 0.4-0.5 because the model is well trained.All the example generations were done on the same seed, indicating good training.P.S. I didn't describe the man and woman's appearance, because back then, the pose description was tied to a specific hair color or body type. You'll have to provide your own clues.In theory, it would be possible to increase the dataset by 3-5 times with different hair colors/body types, but then the number of steps would have to be increased by 3-5 times, while maintaining the proportion of images per pose, which would most likely lead to unsuccessful training.

KREX

Files with SHA256: c2dc7198d260f5ed09d9373153d0fc46a9e8ba749edea33de7a92f67d341fbf4