easy lora

Easy LoRA

Provides an easy-to-use LoRA training tool by dockerized kohya-ss/sd-scripts.

An example of LoRA output

Requirements

• Linux (such as Ubuntu 22.04 LTS) or Windows
• NVIDIA Container Toolkit
  • https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#installing-on-ubuntu-and-debian
• CUDA compatible GPU

Quickstart

Produces the LoRA model that generates anime-style illustrations of Tohoku Zunko.

docker compose build train
docker compose run download_images
docker compose run train gsdf/Counterfeit-V2.5 zunko

# output/zunko.safetensors will be generated

Generate Images Using LoRA

AUTOMATIC1111

Install

git clone https://github.com/AbdBarho/stable-diffusion-webui-docker
cd stable-diffusion-webui-docker
git reset --hard 1df18b803cda07309507088128c7ab999d04de63
docker compose --profile download up --build
docker compose --profile auto up --build

Usage

1. Copy the trained <this-repository>/output/zunko.safetensors to the data/Lora directory
2. Download Counterfeit-V2.5_fp16.safetensors via gsdf/Counterfeit-V2.5 and put it in the data/StableDiffusion directory
3. Download EasyNegative.safetensors via gsdf/EasyNegative and put it in the data/embeddings directory
4. Access http://localhost:7860/ with Chrome and open the AUTOMATIC1111 UI
5. Select the base model Counterfeit-V2.5_fp16.safetensors
   • Stable Diffusion checkpoint
6. Move to the Settings tab
7. Select the VAE vae-ft-mse-840000-ema-pruned.ckpt
   • Settings > Stable Diffusion > SD VAE
8. Apply settings
9. Move to the txt2img tab
10. Restart the Docker container
    • Enter Ctrl+C to terminate the console
11. Enter the prompt below
    • (best quality, masterpiece:1.2), zunko, 1girl, <lora:zunko:1>
12. Enter the negative prompt below
    • (EasyNegative:1)
13. Generate

Prompt to generate the example image in this README.md

(best quality, masterpiece:1.2), zunko, 1girl, <lora:zunko:1>
Negative prompt: (EasyNegative:1)
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 765595793, Size: 512x512, Model hash: 71e703a0fc, Model: Counterfeit-V2.5_fp16, Denoising strength: 0.7, Version: v1.2.1, Hires upscale: 2, Hires upscaler: Latent

Used embeddings: EasyNegative [119b]

Notice

• The images of Tohoku Zunko are provided as the guideline below
  • https://zunko.jp/guideline.html

Links

• kohya-ss/sd-scripts setup tutorial video (in Japanese)
  • https://www.youtube.com/watch?v=N1tXVR9lplM