WLOP Style Stable Diffusion Heywhale

🚀 Wlop-style Stable Diffusion Weight

This project provides a set of Stable Diffusion models trained with the DreamBooth technique. These models can generate images in the 'wgz style', offering different features and performance based on their versions.

🚀 Quick Start

If you want to use this style, the prompt should be: wgz style.

✨ Features

Model Versions

There are three models in the library: V1, V2, and BETA.

• V1: WOLP.ckpt
  • Advantages: Good generation logic, performs well in text-to-image conversion.
  • Disadvantages: Fewer screen details, less imitation of strokes, and the whites of the eyes tend to turn red.
• V2: wgz v2.ckpt
  • Advantages: More details, better facial portrayal.
  • Disadvantages: Eyeballs tend to generate green, and the text generation logic is not good, resulting in poor usability of text-to-image.
• BETA: wgz beta.ckpt
  • This version has relatively complex training maps and poor logic, but sometimes can obtain satisfactory results in image-to-image conversion.

V2 Output Example

Here is a nice output of V2 made by using the image-to-image method, along with the configuration used:

• Prompt: wgz style,portrait of a beautiful women highly detailed,perfect femine face,Classical oil painting,by masamune shirow, by William-tae Kim
• Negative prompt: old,men,two poeple,Three dimensional facial features,deep shadows on the five senses,tall nose,Eye shadow,[High bridge], [narrow nose],Deep orbital fossa
• Steps: 20
• Sampler: DPM++ 2S a Karras
• CFG scale: 9.5
• Denoising strength: 0.51
• Mask blur: 4

BETA Output Example

Below is a nice output of the BETA version. The abstract color swatches are appealing, but the production rate is not high.

V1 Output Example

Here are some samples generated by V1.

🔧 Technical Details

V1 Training Information

• Training with the Elden ring Style ckpt (SD1.5): eldenring-ckpt
• Training parameters:
  • Steps: 12500 steps
  • Learning rate: 1e-6
  • Initial images number: about 100
  • Class image number: 200

📚 Documentation

There are more pictures in the image folder.

💡 Usage Tip

I've tried hard to achieve the desired painting style using Stable Diffusion, with a mix of failures and successes. With my current technology, it's hard to make progress now. So this model will not be updated in the short term. If you have ideas or want to improve together, you can contact me.

• Telegram: https://t.me/hhhyxnh
• QQ: 1602821649

I hope you enjoy this model and create satisfactory works.