Presenting a highly detailed LoRA for Remilia Scarlet, the eternal mistress of the Scarlet Devil Mansion. This model is meticulously trained for SDXL to generate high-quality, "illustrious" anime-style artwork.
This LoRA is designed to faithfully reproduce Remilia's iconic attributes, including her piercing red eyes, characteristic bat wings, and her classic ensemble of a white dress, red ribbon, and frilly mob cap.
A key feature of this model is its versatility in depicting her hair. Reflecting the variety seen across fan-art and official works, you can easily guide the model to generate her with either blue hair or purple hair, granting you greater creative freedom.
Trained on a high-resolution dataset with dynamic bucketing (ARB), this LoRA excels at generating large, detailed images with crisp lines and minimal artifacts, especially in portrait-style aspect ratios (like 768x1024).
How to Use:
Trigger Word: remiliascarlet
Recommended Weight: 0.7 to 1.0. Start around 0.8 for a strong and balanced effect.
Helper Tags: For best results, combine the trigger word with descriptive tags.
Essential: red eyes, bat wings
Appearance: blue hair (or purple hair), mob cap, white dress, red ribbon
Style: anime style, digital art, detailed shading
Note: This is an SDXL LoRA. It is not compatible with SD 1.5 based models.
Description
Resolution: Trained on high-resolution images (832x1216) with Aspect Ratio Bucketing (ARB) enabled, ensuring great performance on taller images and preserving details.
Learning & Optimization: Utilized a differential learning rate for the Unet and Text Encoder, paired with a cosine_with_restarts scheduler. This approach helps the model learn the character's visual concepts accurately while maintaining excellent responsiveness to text prompts.
Network: Built with a network dimension (rank) of 16 and an alpha of 8, offering a good balance between model expressiveness and file size.
FAQ
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.