Try size: 512 x 768 (SD 1.5) / 768 x 1024 (Pony and ILXL)
Weights: 1
Clip skip: 2
Training material from bing and danbooru
Pictures from CalicoMix 8.5 and CalicomixPonyXL_v10 and CalicomixEroill_v50
Note: It is impossible to use danbooru tags completely when training SD 1.5 LoRA, so adjustments must be made, so I used "bathtub"! It’s a bathtub! I added a bathtub! (SD 1.5)
Note: When the girl facing to right, the hand height will be wrong more often, but sometimes it will be normal. (Pony)
Note: Because Illustrious XL v0.1 reacts too strongly to the built-in "goldfish scooping", the training material is difficult for LoRA to learn. So I switched to Anyillustrious XL_v0.1.1 for training, but the prompting effect became weaker. The weakening is just that there is less of a sense of surprise, closer to the feeling of using Pony before. (ILXL)
