Aesthetics Mixed

Aesthetics Mixed is a lora trained with selected aesthetic images. It is meant to improve the aesthetic for base model.

This is a very experimental model for now.

I am still finding the best settings for training a lumina lora, and I will upload the better version in the future.

Features

• Base on Neta Lumina Beta-0624 aes-expereimeent. /model/1612109/

• Many high-quality selected illustrations.

• Improve the aesthetic.

• Improve follow instructions.

• Danbooru tags.

To Do

• Stable version.

• Natural language prompt.

• Training with regularize images for more flexible.

• Training with preference optimization for better aesthetic and more in line with human preferences.

My RTX 3080 is dead after hard work for a long time. I need to attend its funeral, so I may need some time before I can continue.

Next Plans (When Neta Lumina is complete. Or others full fine-tuning model with danbooru tags.)

1. More loras of character.

2. More loras of style.

3. DreamBooth model.

4. Full fine-tuning model.

After some testing, I think that maybe I can train some realistic models too?

SampleHOW TO USE

1. Download Base model. /model/1612109/

2. Download lora file.

3. Move to path: ComfyUI/models/loras/

Prompt

You are an assistant designed to generate anime images based on textual prompts.<Prompt Start>aes-mixed, masterpiece, best quality,

Negative Prompt

You are an assistant designed to generate anime images based on textual prompts. <Prompt Start> 3d, realistic, blurry, worst quality, low quality, normal quality, freckle, disabled body, sketches, manicure, lowres, bad anatomy, bad hands, text, error, bad lamb, bad finger, extra fingers, missing fingers, extra digit, fewer digits, cropped, jpeg artifacts, signature, watermark, username, company name, 

• steps: 28

• cfg: 5

• sampler_name: res_multistep

• scheduler: sgm_uniform