Trained LoRA for Akutox3's art style on SDXL. Built with 67 curated images over 150 epochs.

Works best at 1.0 strength, but can use 0.6 - 0.8 for subtle influence. Trigger: "Akutox3"

Training:

- 67 images × 150 epochs

- Dim/Conv: 32, Alpha: 16

- AdamW8bit + Huber loss

- Cosine w/restarts scheduler

- Full augmentations

- Base: noobAI Epsilon 1.0

Technical Observations

Compatibility & Style

• Functions well with basic implementations without interference

• Blends with other art styles, though consistency of style influence varies

Known Limitations

• Weapon Rendering:

  • Good performance with swords

  • Poor knife rendering - note that this may be due to common limitations across many art styles rather than specific to Akutox LoRA

  • Results may vary; your experience could differ

• Pose Constraints:

  • Tendency to default to sitting positions

  • Particularly noticeable with prompts involving animals on arms

  • May be related to either LoRA or base model behavior (unconfirmed)

  • Individual results may vary depending on specific prompts and settings

Integration Notes

• Successfully maintains basic functionality

• Adaptable with other styles while preserving core characteristics

• Note: All observations are based on specific testing conditions and may not represent all possible use cases

Enjoy!

描述:

训练词语: