Overview

### 📁 Technical Archive & Mirror

For detailed merge logic (Differential Summation), technical scripts, and high-speed mirrors, please visit the official FjikoAI technical archive on Hugging Face:

SAKURA_Pony_v0.3 - Hugging Face Repository

https://huggingface.co/FjikoAI/SAKURA_Pony_v0.3

---

SAKURA_Pony_v0.3 is a high-quality model based on the Pony Diffusion V6 XL architecture.

Building upon the "SAKURA" series' philosophy of clean lighting and refined aesthetics, this version is specifically tuned for a balanced blend of realistic textures and characteristic Pony-style versatility.

Technical Details (Differential Weighted Merge)

This model was created using a Differential Merge technique (Delta Merging). Rather than a simple weighted sum, I used a custom script to calculate the "essence" (differences) of multiple specialized models relative to the SDXL Base 1.0, and then recomposed them into a unified checkpoint.

Recommended Settings

To get the best results with SAKURA_Pony_v0.3, please use the following settings:

Base Model: SDXL 1.0 / Pony

Sampler: Euler

Steps: 30

CFG Scale: 7.5 (Pony standard)

Clip Skip: 2

VAE: sdxl_vae.safetensors (or any SDXL-compatible VAE)