“Luna Body – Hyperreal SDXL v1” is a character-focused LoRA trained on a small but highly curated dataset designed to reproduce the realistic body structure, muscle definition, and anatomical proportions of a tall, athletic woman under controlled lighting and neutral environments.
This model captures realistic curvature, light reflection on latex-like materials, and accurate muscle tone transitions, suitable for projects that aim to render hyperrealistic female anatomy with an emphasis on natural balance, posture, and proportional volume.
The LoRA was trained using SDXL Standard Base with caption-based annotations at 1024px resolution, focusing on body detail, surface gloss, and photometric consistency between different outfits and lighting conditions.
It generalizes well with most hyperreal or semi-realistic checkpoints, especially those emphasizing cinematic lighting and detailed skin rendering.
Best used for:
Hyperreal female portraits
Full-body character renders
Training foundations for stylized or anatomical realism
Testing proportional consistency across camera angles
Not recommended for:
Cartoon or anime bases (may distort anatomy)
Overexposed or extreme lighting conditions
Technical parameters:
Base model: SDXL v1 Standard
9 captioned images, 10 epochs, resolution 1024px
Trained using cosine LR scheduler with restarts, network dim 32 / alpha 16
Ideal strength range: 0.6–0.8 (for proportional realism)
Description
Luna Body – Hyperreal SDXL v2 (SharpEdition)
This update refines the model’s rendering consistency and texture precision while maintaining anatomical proportions identical to version 1.
Improvements include:
• Enhanced edge sharpness and contour fidelity.
• Better surface lighting balance and color uniformity.
• Reduced blur artifacts on close-up compositions.
• Maintained compatibility with the same trigger word: luna_body.
The dataset used for this version is identical in subject scope but reprocessed for clarity, contrast, and uniform size (1024×1500).
Ideal for users seeking cleaner close-up renders and more controlled highlights without altering the original model’s structure or tone.