This model is a high-performance hybrid merge based on Illustrious-XL, specifically tuned to bridge the gap between Illustrious's superior prompt understanding and Pony Diffusion's aesthetic character sculpting.
Version 1.6 marks a significant milestone in achieving a tangible "evolution of form," ensuring that visual changes are clearly visible even when using the same prompt and seed.
🛠 Technical Features: Triple Hybrid Merge
Unlike standard weighted merges that often result in "blurry" details, this model was constructed using a custom script implementing three core logics:
True Add Difference (ALPHA=0.25):
Maintains the integrity of the Illustrious base while injecting character depth by adding the specific difference between Pony and the SDXL base.
CLIP Protection:
The Text Encoders (CLIP/Conditioner) were excluded from the merge. This ensures 100% compatibility with Illustrious-specific tags and maintains high responsiveness to complex prompts.
Statistical Rescale Control:
Each layer's standard deviation was mathematically recalibrated to match the base model’s range, effectively suppressing "color bleeding" and "contrast anomalies."
🚀 Recommended Settings
Sampler: Euler
Steps: 20 - 30
CFG Scale: 3.5 - 5.0|
✒️ Developer's Postscript (Self-Analysis)
This version focuses on achieving a solid "Checkpoint" for visual character evolution and prompt reliability.
Successes:
Clearer facial contours, improved nose/lip definition, and higher-density eye detailing.
Future Challenges:
Oriental Aesthetics: While the goal was to lean further into "Eastern/Asian" facial features, I prioritized contour depth and stability in this version.
Color Vibrancy: Some of the signature "SAKURA" color intensity was traded for improved sculptural detail. Re-balancing this vibrancy is the primary goal for the next major update.
