MIKAN_Pony_v0.9

MIKAN Pony v0.9

<img src="https://cdn-uploads.huggingface.co/production/uploads/68d789141aa476c348609d17/08HqET-T7jJYH-eEKKMiW.png" width="600">

1. Concept: "Balanced East Asian Realism"

MIKAN_Pony_v0.9 is a custom Pony Diffusion V6 XL merge model designed to achieve a balanced combination of photorealistic rendering, East Asian facial aesthetics, and strong prompt responsiveness.

Built using a Differential Merge Framework, the model preserves the compositional flexibility and expressive capabilities of Pony while introducing enhanced skin rendering, refined facial structures, and improved lighting realism.

The primary objective of MIKAN_Pony is to provide stable generation of realistic East Asian characters without sacrificing the creative versatility that has made Pony Diffusion popular among creators.

---

2. Technical Methodology

This model was constructed using a custom differential merge script based on SDXL Base 1.0 as a neutral anchor model.

Instead of performing a traditional weighted merge, MIKAN_Pony calculates the differential contribution of each source model relative to SDXL Base and then reconstructs a new model through weighted delta summation.

Merge Composition

Anchor Model

• SDXL Base 1.0

Primary Base Model

• Pony Diffusion V6 XL

Differential Weights

• waiREALMIX_v11.safetensors : 40%
• 5moonPonyAsian_v10.safetensors : 25%
• ponyLoraLover_v10.safetensors : 15%
• waiDOLLCN_v40.safetensors : 10%
• tsubakiMix_v15.safetensors : 5%
• unholyDesireMixFoolS_v60.safetensors : 5%

Differential Merge Logic

Result = SDXL_Base + Σ((Model_n - SDXL_Base) × Weight_n)

By reconstructing the final model from weighted deltas rather than direct parameter averaging, MIKAN_Pony aims to preserve desirable traits from each source model while reducing unwanted cross-model interference.

---

3. Expected Visual Characteristics

Strengths

• Realistic East Asian facial features
• Improved skin texture and material rendering
• Balanced photorealistic output
• Natural lighting response
• Strong prompt adherence
• Stable composition generation
• Retains Pony's expressive character capabilities

Recommended Use Cases

• Portrait photography
• Fashion photography
• Lifestyle imagery
• Idol-themed artwork
• Realistic character illustrations
• PBR-oriented image generation

---

4. Recommended Settings

Base Workflow

• Sampler: DPM++ 2M Karras
• Steps: 25–35
• CFG Scale: 5.0–7.0
• Resolution: 1024×1024 or higher

Suggested Negative Prompt

(score_4, score_5, score_6, censored, low quality, worst quality, blurry, 3d, render, cartoon)

---

5. Credits

Base Model

• Pony Diffusion V6 XL

Source Models

• waiREALMIX_v11
• 5moonPonyAsian_v10
• ponyLoraLover_v10
• waiDOLLCN_v40
• tsubakiMix_v15
• unholyDesireMixFoolS_v60

Special thanks to the original creators of the source models used in this merge.

---

6. License

This model is based on Pony Diffusion V6 XL and follows the applicable licensing terms of the original base model and source models.

Please review the license requirements of all upstream models before commercial or public redistribution.

---

🌟 Join the Community

<div align="center"> <p>If you enjoy this model, please consider sharing your work and leaving a review on Civitai!</p> <a href="https://civitai.com/models/2856699/mikanponyv09?modelVersionId=3226522"> <img src="https://img.shields.io/badge/Civitai-Model_Page-blue?style=for-the-badge&logo=civitai" alt="Civitai"> </a> </div>

---

About FjikoAI

FjikoAI focuses on the fusion of logical technical control and emotional aesthetics.

By combining differential model engineering, physically based rendering principles, and practical image-generation workflows, FjikoAI aims to create models that balance artistic freedom with technical reliability.