Hexagon Anime Style

Hexagon Anime Style LoRA is designed to provide a highly versatile and generalized anime aesthetic. It is specifically trained for Pony SDXL models and demonstrates excellent compatibility with most models and other LoRAs. This LoRA aims to fit seamlessly into a variety of workflows and creative projects, making it a reliable choice for diverse use cases. It features both a 256, 768 and a 1024 network configuration, available as different versions in this post. While I’ve been working to reduce its size below 2GB, optimization for a smaller version is ongoing.

This LoRA supports both standard and iPhone wallpaper aspect ratios. Its style blends the nostalgic charm of early 2000s anime with modern aesthetics, while maintaining a clean and simple approach. The Christmas theme used for the examples highlights its flexibility, but it performs equally well across other themes and genres.

The name “Hexagon Anime Style” reflects its ability to integrate smoothly with other models and LoRAs, providing a well-rounded, adaptable style that complements a wide range of projects.

Performance and Compatibility:

The outputs vary depending on the base model used, especially with finer details like hands, eyes, and backgrounds. I’ve tested this LoRA with various other models and achieved consistently good results. Its generalized nature ensures compatibility with a wide variety of models and themes, making it a strong choice for both SFW and NSFW content (with stronger results for NSFW).

All example images in this post are Christmas-themed, used as a demonstration of its versatility. However, this LoRA is not limited to holiday themes and can work effectively across a wide range of styles and subjects.

Models Used in Examples:

• geekpowerCheckpoints_celShade.safetensors

• prefectPonyXL_v10.safetensors

• hassakuXLHentai_v13.safetensors

Workflow Suggestions:

General Workflow (1024x1536):

1. Initial Generation Resolution: 1024x1536

2. Post-Processing: Upscale 1.5x with inpainting at 0.45 denoising (repeat twice).

3. Final Output Resolution: 2304x3456

iPhone Workflow (880x1912 using Tiled Diffusion):

1. Initial Generation Resolution: 880x1912

2. Tiled Diffusion Upscaling:

• Use tiled diffusion to split the image into smaller tiles (e.g., 128x128 or 256x256). This helps reduce VRAM usage while maintaining high-quality results.

• Apply 2x upscaling with tiled diffusion (e.g., using tools like Tiled Diffusion in Automatic1111 or ComfyUI).

• Set denoising strength to around 0.3–0.5 for detail enhancement without over-blurring.

3. Final Resolution: 1760x3824

4. Optionally, use inpainting for additional refinements, particularly on complex details like hands, eyes, and backgrounds.

Key Notes:

• All example images provided were raw, first-generation outputs at a resolution of 1024x1536 without any post-processing.

• This LoRA excels when images are refined through upscaling, inpainting, or further adjustments with complementary models.

• The “Hexagon Anime Style” LoRA is designed to fit seamlessly with other models and LoRAs, making it a reliable choice for users looking for a generalized style that works across different themes.

• Tiled diffusion is especially useful for devices with limited VRAM, allowing high-resolution upscaling without exceeding memory constraints.