Experiments with Stable Diffusion

About

This repository is a fork of Justin Pinkney's stable diffusion repository.

It presents the use case of fine-tuning a text2image stable diffusion model with a BLIP captioned naruto face dataset.
In this case, we generate Naruto style images from a text prompt:

"Bill Gates with a hoodie", "John Oliver with Naruto style", "Hello Kitty with Naruto style", "Lebron James with a hat", "Mickael Jackson as a ninja", "Banksy Street art of ninja"

For a step by step guide on training the stable diffusion model see the Lambda Labs examples repo.

Usage

!pip install diffusers==0.3.0
!pip install transformers scipy ftfy

import torch
from diffusers import StableDiffusionPipeline
from torch import autocast

pipe = StableDiffusionPipeline.from_pretrained("lambdalabs/sd-naruto-diffusers", torch_dtype=torch.float16)  
pipe = pipe.to("cuda")

prompt = "Yoda"
scale = 7.5
n_samples = 2

# Sometimes the nsfw checker is confused by the Naruto images, you can disable
# it at your own risk here
disable_safety = False

if disable_safety:
  def null_safety(images, **kwargs):
      return images, False
  pipe.safety_checker = null_safety

with autocast("cuda"):
  images = pipe(n_samples*[prompt], guidance_scale=scale).images

for idx, im in enumerate(images):
  im.save(f"{idx:06}.png")