Image-to-Image using Stable Diffusion 1.5

# @title Install dependencies
!uv pip install -q diffusers["torch"]==0.35.1 transformers==4.56.2 accelerate==1.10.1

Load SD 1.5 model using pipeline

import torch
from diffusers import StableDiffusionImg2ImgPipeline

# Load the img2img pipeline
print("Loading Stable Diffusion img2img pipeline...")
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    dtype=torch.float16
)
pipe = pipe.to("cuda")

Loading Stable Diffusion img2img pipeline...

Keyword arguments {'dtype': torch.float16} are not expected by StableDiffusionImg2ImgPipeline and will be ignored.

Generate Intermediate Images

from PIL import Image
import requests
from io import BytesIO

IMAGE_URL = "https://raw.githubusercontent.com/simonguest/CS-394/refs/heads/main/src/04/images/luna.jpg"
SEED = 128763
PROMPT = "a goldendoodle wearing sunglasses, high quality, detailed"
NEGATIVE_PROMPT = "blurry, low quality, distorted"

response = requests.get(IMAGE_URL)
init_image = Image.open(BytesIO(response.content)).convert("RGB")

# Resize to standard size for faster processing
init_image = init_image.resize((512, 712))

display(init_image)

Function to generate using strength param

def generate_image(strength):
  return pipe(
      prompt=PROMPT,
      negative_prompt=NEGATIVE_PROMPT,
      image=init_image,
      strength=strength,
      guidance_scale=7.5,
      num_inference_steps=30,
      generator=torch.Generator().manual_seed(SEED) if SEED != -1 else None,
  ).images[0]

Strength 0.3

display(generate_image(0.3))

Strength 0.5

display(generate_image(0.5))

Strength 0.7

display(generate_image(0.7))

Strength 0.9

display(generate_image(0.9))