HelioBasev5.safetensors Model Card

This model was created by @MoonShoesMel.

Model Description

The model is a combination of 24 custom-trained Stable Diffusion 1.5 models, trained in DreamBooth on various styles and concepts from my own artwork and photography.

Data

Each model used in the merge was trained using 400 images, paired with a caption set, using the following parameters:

• 40000 Training Steps
• Learning Rate: 7e-6
• Text Encoder Training Steps: 4000
• Text Encoder Learning Rate: 1e-6
• Resolution: 512

Performance

Each model that was used in the merge was tested every 4000 steps. The merging process used a 50/50 approach and involved over 200 merges, followed by fine-tuning.

Usage

The model excels in generating ethereal Kawaii art-style images with slight realism, specifically anime-style images of female characters. It has been trained to correct hands.

Example Inputs and Outputs

• Prompts: (kawaii ethereal ghost woman with big, adorable eyes floats playfully in the air), dark and moonlit forest backdrop, spooky and cute, twisted trees and creeping shadows, Glowing jack-o-lanterns, (realism, finely detailed)
• Negative Prompt: blurry, low quality, low-res, watermark, out of frame, dark, overly saturated, (black background, empty background, simplistic, minimalist:1.3)
• Sampling Steps: 30
• Sampling Method: DPM++ 2S a Karras
• CFG Scale: 7

Limitations

The model does not handle images of animals or men very well and may lack diversity in representations of people.

Requirements

This is a safetensors checkpoint for StableDiffusion 1.5.

Guidelines

Please do not use for NSFW content. I've done my best to remove most NSFW concepts, however some may still remain.

Contact

You can reach out to me on Twitter at @MoonShoesMel.