Helluva Boss Style [NoobAI XL / V-Pred]

This LoRA captures the distinct art style of Helluva Boss. It was trained specifically on NoobAI XL vPred 1.0, meaning it is designed for V-Prediction models.

⚠️ Important: This is a V-Pred LoRA. For best results, use it with NoobAI XL or other V-Prediction based SDXL checkpoints (e.g., Chenkin). Using it on standard Epsilon-prediction models (like base SDXL or older Pony versions) or with incorrect settings will result in washed-out colors or "fried" images.

Usage & Recommended Settings

Trigger Word: hazbinhotel
Strength: 0.7 - 1.0 (Trained with high Alpha, start lower if it over-bakes).
Sampler: Euler A CFG++ (Essential: Standard Euler A may cause greying/desaturation).
Scheduler: SGM Uniform
CFG Scale: 1.4
SwarmUI Users: Do not add the <lora:hazbinstyle> tag to your prompt manually if you have selected it in the LoRA tab. This will double-load the weights and ruin the image.

Behavior & Quirks

Colored Sclera: True to the source material, this model has a strong tendency to generate colored scleras (the whites of the eyes), even when not prompted.
"Eye" Hallucinations: The model may occasionally interpret glowing objects (like robotic parts or background lights) as eyes/pupils.
Furry Bias: Because many characters in the training data are anthropomorphic, the model may add furry traits (snouts, fur texture) to characters that have animal ears or tails.
Character Overwrite: The style applies easily to OCs. However, for established characters (e.g., from other anime), the model may struggle to fully override the character's original art style depending on the checkpoint used.

Training Data

Resolution: 1024x1024
Source: 60 hand-picked images of the show's style.
Method: Trained on the UNet only using the CAME optimizer.

Helluva Boss - Hazbin Hotel Style - V1