Model Overview

This SDXL anime model is crafted with passion and precision by a single developer. Designed to capture a clean, vivid aesthetic, it delivers high-tier generation quality comparable to NovelAI Diffusion V3 (and V4).

The training dataset was fully curated, filtered, and processed manually—combining top-rated Danbooru posts, high-quality NovelAI generations, and hand-picked anime screenshots to eliminate noise and fluff.

Recommended Resolutions

The model is optimized and performs best at native SDXL aspect ratios:

• 1216 × 832 (Landscape)

• 832 × 1216 (Portrait)

• 1024 × 1024 (Square)

Recommended Settings

• Steps: 24 – 28

• CFG Scale: 3.0 – 6.0

• Hires.fix: Optional (the model outputs crisp, high-detail images natively, but feel free to use it if you want extra upscaling).

Available Versions & Updates

• Epoch 2: Available right here on Civitai for free! Enjoy testing and generation.

• Epoch 3: Currently undergoing internal testing and fine-tuning to deliver even better detail, composition, and prompt stability.

P.S. Early build notice: dataset structuring for text encoders is still missing in this version, and I'm active testing variable repeats for less common styles and characters. I'm working on a massive project right now — follow for future updates and the polished final release!