A LoRA focused on bowlegged pose for Anima, trained on Anima Base v1.0 with Citron Legacy.

Trained with 474 images, 10 epochs, and 15400 total training steps.

Trigger word: bowlegged pose

Tips:

• legs apart gives the regular bowlegged stance.

• wide spread legs pushes the legs much farther apart.

• low stance lowers the center of gravity without turning it into a full squat.

• feet turned outward and on tiptoes work great depending on the pose.

• Works nicely with different camera angles, clothing, and art styles.

• Also supports sitting, leaning, back views, prone, supine, and even some interaction poses.

• I intentionally included a lot of pose variations instead of only straight standing references, so it should stay pretty flexible instead of locking itself into one exact composition.

Hope you enjoy it!

Keep gooning y'all~

Training Data ↓

Base Model: Anima Base v1.0
Network Dim: 20
Network Alpha: 20
Learning Rate: 1e-4
Optimizer: AdamW8bit
Resolution: 768
Batch Size: 1
Epochs: 10
Steps: 15400
Caption Dropout: 0.1
Mixed Precision: bf16

Dataset:

• 474 manually curated images

• 1540 effective images per epoch

• 15400 total training steps

• Multiple pose variations (standing, low stance, sitting, leaning, back view, prone, supine, interactions and 3D references)