Update Logs

• 2023/10/1: Adjust image tags to make the descriptions of training images more accurate.

Introduction

A LORA model of cyberpunk style. Since cyberpunk is a grand concept and has numerous related works, I only select some distinctive and beautiful images as the dataset.

Usage

• Trigger tag: cyberpunk, science fiction, neon lights

• Key Feature:

  • For scenery: scenery, skyscraper, city lights, cityscape, outdoors, city, building, holographic interface, hologram, star \(sky\)

  • For human: cyborg, robot, mecha, mecha musume, mechanical arms, jacket, mask, smoke, respirator, hologram, asymmetrical hair

• This LORA can be used in conjunction with other LORAs to generate masterpieces.

Generating Settings

• Base Model: Arbitrary anime model (realistic model may be also supported)

• Sampler: DPM++2M Karras, DPM++3M Karras and other DPM samplers are recommended.

• Eta noise delta: 31337 or 0

• Upscale Method: HiresFix+4xAnimeSharp

• Max Resolution without Upscaling: 768x768

• VAE: kl-f8-anime2 is recommended.

• LORA Weights: 0.65~0.8 are recommended.

Cautions

• For Anime model, pls disable 'face restoration' during generating.

• Make sure set this option in 'Stable Diffusion' settings to 'CPU' to successfully regenerate the preview images with the same seed.

• The training resolution of this model is 768x768, so generating resolution must not exceed this range. Try to use 'hires fix' to prevent this problem and help to generate higher-resolution images.