Sign In

Aika - Eternal Arcadia LoRA

62
356
618
4
Verified:
SafeTensor
Type
LoRA
Stats
83
534
273
Reviews
Published
Dec 13, 2024
Base Model
Illustrious
Training
Steps: 20,000
Epochs: 77
Usage Tips
Clip Skip: 1
Strength: 0.6
Trigger Words
aika
aika (eternal arcadia)
orange hair
brown eyes
low twintails
twin braids
goggles
single glove
Hash
AutoV2
38A6301240

=== SDXL VERSION (Illustrious) ===

Description: Initial release of Aika character style LoRA for SDXL, based on thorough testing of the character from Eternal Arcadia. Includes comprehensive stability checks across multiple weights and epochs.

Latest Changes v1.0:

  • Verified performance across multiple weights

  • Tested with different art styles and resolutions

  • Established behavioral patterns and limitations

  • Optimal configuration determined through testing

Required Base Information:

  • Trigger word: Aika

  • Recommended weight range: 0.4-0.6 (best results at 0.6)

  • Optimal epoch: 77

Known Behaviors:

  • Stable character generation within recommended weight range

  • Works well with wider resolutions

  • Good art style compatibility

  • Minimal model interference at lower weights

  • Default costume shows high persistence

  • Random object generation in hands

  • Weapon prompts may create artifacts

  • Handgun prompts show contamination

  • Gloves and blue sweatband appear/disappear randomly

  • Twin pigtails occasionally generate as three instead of two

Tips:

  1. Stay within recommended weight range for best results

  2. Avoid weapon-related prompts to minimize artifacts

  3. Expect and work around random hand object generation

  4. Be prepared for accessory inconsistencies

  5. Default costume elements will tend to persist

=== SD 1.5 VERSION ===

Creator's Description: Hi all! This is a LoRA requested at a bounty for Aika from Eternal Arcadia!

Technical Details:

  • Trigger word: Aikostyle (Note: This is the correct trigger for SD 1.5 version)

Both versions have been thoroughly tested and optimized for their respective models. Choose the version that matches your base model for optimal results.