タイプ
Checkpoint
投稿日時
2025-04-08
基本モデル
Illustrious
Model Parameters
clip_skip：1
トレーニングパラメータ
Epochs：100
Steps：0
バージョン紹介
This version support native 1080p Generation Best resolution 1560x1920 All pew. generated in native hd
🧠 Key Features
- Max Native Resolution: 1560x1920 (Tested in Forge / A1111)
- Full XL LoRA Compatibility: Works seamlessly with all XL LoRA models (e.g., Pony, Anime styles, etc.)
- Special Scoring System:score_SFW / halfSFW / ????: 0–2.0score_feeling_somefeeling: 1–2.0 (Face detail & emotion quality)score_details: 0–2.0 (Overall image detailing)score_fullhdanatomyfix: 0 or 1 (Fixes anatomical issues in HD)
- Built-in Detail Enhancement Engine
- Fixed Clip Integration – Special internal tweaks for better prompt coherence
- Next-Gen Sampler Support – Compatible with the newest sampling methods
- Custom Trained VAE – Specifically trained to match the model’s aesthetic
🐞 Known Issues (Transparency First!)
- Score Tags: May appear on some images during generation.
- Background Artefacts: Occasionally, background characters may appear incorrectly as pasted images or ghosted figures.
- Hires.fix Scaling Bug: Scaling from native 1080p to 4K may break image structure. This issue stems from mismatches between specified and actual generation dimensions — fix in progress!
