Cheon Geuk | Character LoRA (Illustrious-XL)
This is a dedicated character LoRA for Cheon Geuk from the manhwa The Former Supreme / Jeonjikjijon.
To ensure the highest possible quality and composition versatility, this model was trained on a highly curated dataset consisting of 31 high-resolution, high-detail images optimized natively for the Illustrious-XL architecture. No specific backgrounds, rigid poses, or static aesthetics are baked into the weights, ensuring maximum prompt responsiveness.
⚠️ Work In Progress Note:
This model is an ongoing project. It will be continuously updated version-by-version if and when more high-detailed images become available for training.
Key Capabilities:
High Fidelity: Accurately reproduces her signature character design, including her facial structure, eye shape, and distinct long hair configuration.
Exceptional Flexibility: Highly receptive to clothing modifications, environment changes, and complex compositional prompts.
Versatile Application: The model functions seamlessly across various workflows, adapting effortlessly to traditional martial arts settings, modern apparel shifts, or specialized content requests based entirely on user prompting.
Generation Recommendations:
Base Model: Formulated for Illustrious XL and derivative community models.
LoRA Weight:
0.8 - 1.2(Adjust according to your specific checkpoint and prompt complexity).
Feedback and community generations are highly encouraged. Please consider uploading your renders to the review section below to help optimize future version updates.
Description
v1.0 — Initial Release (Illustrious-XL)
What’s Included in This Version:
High-Fidelity Aesthetics: Accurately captures Cheon-Geuk’s signature manhwa look, including her sharp eyes, high-waist split robe, red patterned lapels, bamboo hat with notched brim, and sheer reddish veil.
Smart Gear Unbinding: Trained with a dual natural language (JoyCaption) + Danbooru tag pipeline, allowing you to easily unmask her or prompt alternative outfits without forcing her default assassin hat/mask.
Optimized for Illustrious-XL: Built from the ground up for Illustrious-XL base models for crisp line art, strong tag adherence, and clean color rendering.
Training Dataset & Technical Summary:
Dataset Size: 36 hand-selected, high-resolution manhwa crops.
Total Steps: 3,240 total steps (15 repeats × 6 epochs).
Captioning: JoyCaption (4-bit NF4) natural language descriptions + BooruDatasetTagManager (WD v3 Swinv2) auto-tagging.
Precision/Format: Trained in
bf16, outputted as.safetensors.
Usage & Setup Tips:
Trigger Tag:
cheon_geukRecommended Weight:
0.8–1.0ADetailer Settings: For face repairs, keep denoise low (
0.28–0.32) and make sure "Append main prompt LoRAs" is checked so her sheer face mask renders smoothly without smudging.



















