stable-diffusion text-to-image safetensors

Holicon

A series of models based on my personal preferences.

Holicon has no actual meaning. Suffix may represent a certain artist or the abbreviation of the models used in merge. All models listed are baked in VAE.

Tips

  1. Please feel free to use whatever parameters you want.

  2. If you want to get images similar to the examples images or want to know some of my usage of these models, you can refer to introduction.

  3. Download the image with a suffix other than 0 to read the parameters.

Introduction

  1. Holicon-F79

Suitable for generating low saturation and more flat images, but it is weaker than other models in terms of prompt response.

F79_0

  1. Holicon-Hiten

Suitable for generating my personal favorite character with slightly juvenile and whiter skin, thanks to the merging of PVC type models.

Hiten_0

  1. Holicon-mao

This model can generate characters similar to Holicon-Hiten, but the image will be more pinkish. It has better scene generation capabilities. In terms of usage, I recommend using it only with Hires fix. (Automatic1111-stable-diffusion-webui 1.6.0+)

mao_0

Examples used on Hires fix.

Nordrin_little v2.5 as the first pass and Holicon-mao as Hires fix.

mao_first_pass_generate_0

Holicon-F79 as the first pass and Holicon-mao as Hires fix.

mao_F79_4_second_hires_fix_0

Recommended Settings

Sampler: DPM++ SDE (30 ~ 50 steps)

Hires fix Sampler: Euler a (15 ~ 40 steps)

Upscaler 1: Latent (bicubic antialiased) for more details

Upscaler 2: ScuNET PSNR for cleaner results

Denoising: 0.5 ~ 0.6

CFG: 7 ~ 9