mapchipLora: A LoRA Model for SDXL

mapchipLora is a LoRA (Low-Rank Adaptation) model designed for Stable Diffusion XL (SDXL) v1.0. It specializes in generating images with a tilechip aesthetic, commonly used in pixel art and game design.

Model Overview

• Base Model: SDXL v1.0
• Training Data: Images originally at resolutions of 16×16, 32×32, and 48×48 pixels, but enlarged to 1024×1024 using the nearest neighbor method.
• Trigger Words: mapchip, followed by 1616, 3232, or 4848 to specify the desired pixel resolution

Usage Instructions

1. Base Model: Ensure that SDXL v1.0 is loaded as the base model.

2. Prompt Structure:

   • Use the following format to generate images:

     mapchip, {1616 or 3232 or 4848}, {your prompt}, <lora:mapchipLora:strength>
     

   • Replace {your prompt} with your desired image description.
   • Choose 1616, 3232, or 4848 based on the desired pixel resolution.
   • For example:

     mapchip, 3232, forest landscape, <lora:mapchipLora:0.7>
     

3. LoRA Strength:

   • Set the LoRA strength between 0.7 and 0.75 for optimal results.
   • Increasing the strength enhances pixel accuracy but may reduce prompt influence.

4. Image Resolution:

   • Generate images at a resolution of 1024×1024 pixels.
   • Use image editing software like Aseprite to downscale the image using the nearest neighbor method and adjust color counts as needed.

5. Negative Prompts:

   • The effect of negative prompts is uncertain.
   • Tilechip-style images can be generated effectively without them.

6. img2img Usage:

   • Applying mapchipLora to existing texture images using img2img is also effective.

License

This model is licensed under the Apache 2.0 License.

Tags

• LoRA
• SDXL