Anime-style Stable Diffusion!

Ever wanted to generate images of a particular style with the current state of the art stable diffusion models? With this repo, you can generate all types of styles.

Girl with bunny ears

A man wearing a mask

A powerful man with fire around his body.

An old portrait of a woman

A girl playing with a dog

Two boys celebrating a soccer goal

A portrait of a blonde man

A powerful female warrior

Run the model locally:

• Install dependencies

git clone https://github.com/RSDP101/Anime-SD.git
pip install virtualenv
python -m venv venv
source venv/bin/activate
cd Anime-SD/examples/text_to_image/
pip install git+https://github.com/huggingface/diffusers.git
pip install -U -r requirements.txt
accelerate config default

• Run training on a SD model and dataset. The default ones are 'CompVis/stable-diffusion-v1-4' and rod101/Anime1K.

# Change model and dataset configuration at the training.py script.
python3 training.py

Training will save model's weight to sd-anime-model.

• Run inference on your fine-tuned model.

python3 inference.py

You can change model and dataset configuration at the training.py script. Pass the prompt you want to generate at the inference.py script.

Enjoy!