Senko-san-Illustrious-XL-v.01 - v2.0 Showcase

Update - Complete Retrain

This version is a complete rebuild of the original Senko-san LoRA.

• Expanded the training dataset from 118 to 598 captioned images

• Retrained on Illustrious XL v0.1 using ostris/ai-toolkit

• Increased training to 2,000 steps

• Updated the network to rank 32 / alpha 32

• Trained both the UNet and text encoders

• Improved character likeness, outfit flexibility, and scene adaptability

• Produces clean results at full LoRA strength without visible overbaking

• No large negative prompt is required
  

The trigger word remains:

  senko_san                                                                                

For this version, use a LoRA weight of 0.85–1.0:

• 0.85 for greater clothing and scene flexibility

• 1.0 for maximum character likeness

Recommended starting settings: Euler a, 28 steps, CFG 7, Clip Skip 2, 1024×1024.