diff --git a/README.md b/README.md
index 7d0a31d..ae9dd45 100644
--- a/README.md
+++ b/README.md
@@ -49,6 +49,19 @@ source venv/bin/activate
 python generate.py "A futuristic cityscape at sunset, highly detailed, cyberpunk style, neon lights"
 ```
 
+### Image-to-Image
+Transform an existing image based on your prompt.
+
+```bash
+python generate.py "A cyberpunk version of this photo" --image "path/to/my_photo.jpg" --strength 0.75
+```
+
+- `--image`: Path to your input image (JPG/PNG).
+- `--strength`: How much to change the image (0.0 to 1.0).
+  - `0.3`: Subtle changes
+  - `0.75`: Default, balanced mix
+  - `1.0`: Completely new image
+
 ### Advanced Options
 You can customize the resolution and quality settings:
 
diff --git a/generate.py b/generate.py
index b2e4f4c..df90105 100644
--- a/generate.py
+++ b/generate.py
@@ -3,12 +3,13 @@ import os
 os.environ["HF_HOME"] = os.path.abspath("hf_cache")
 
 import torch
-from diffusers import StableDiffusionXLPipeline, UNet2DConditionModel, EulerDiscreteScheduler
+from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline, UNet2DConditionModel, EulerDiscreteScheduler
+from diffusers.utils import load_image
 from safetensors.torch import load_file
 from datetime import datetime
 import argparse
 
-def generate_image(prompt, width=1920, height=1080, steps=4):
+def generate_image(prompt, width=1920, height=1080, steps=4, image_path=None, strength=0.75):
     # Setup paths
     # Check if local base model exists, otherwise use repo ID (but we are downloading it locally)
     local_base = os.path.join(os.path.dirname(__file__), "models", "base")
@@ -33,7 +34,12 @@ def generate_image(prompt, width=1920, height=1080, steps=4):
 
     # Load Pipeline
     print("Loading Pipeline...")
-    pipe = StableDiffusionXLPipeline.from_pretrained(base, unet=unet, torch_dtype=torch.float16, variant="fp16").to(device)
+    if image_path:
+        print(f"Initializing Image-to-Image Pipeline with input: {image_path}")
+        pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(base, unet=unet, torch_dtype=torch.float16, variant="fp16").to(device)
+    else:
+        print("Initializing Text-to-Image Pipeline")
+        pipe = StableDiffusionXLPipeline.from_pretrained(base, unet=unet, torch_dtype=torch.float16, variant="fp16").to(device)
     
     # Optimizations for Mac/MPS
     print("Enabling attention slicing for memory efficiency...")
@@ -45,8 +51,15 @@ def generate_image(prompt, width=1920, height=1080, steps=4):
     pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
 
     # Generate
-    print(f"Generating {width}x{height} image for prompt: '{prompt}'")
-    image = pipe(prompt, num_inference_steps=steps, guidance_scale=0, width=width, height=height).images[0]
+    # Generate
+    print(f"Generating image for prompt: '{prompt}'")
+    if image_path:
+        init_image = load_image(image_path).convert("RGB")
+        # Resize input image to target dimensions to avoid size mismatch
+        init_image = init_image.resize((width, height))
+        image = pipe(prompt, image=init_image, strength=strength, num_inference_steps=steps, guidance_scale=0).images[0]
+    else:
+        image = pipe(prompt, num_inference_steps=steps, guidance_scale=0, width=width, height=height).images[0]
 
     # Save
     # Save
@@ -67,7 +80,9 @@ if __name__ == "__main__":
     parser.add_argument("--width", type=int, default=1920, help="Image width (default: 1920)")
     parser.add_argument("--height", type=int, default=1080, help="Image height (default: 1080)")
     parser.add_argument("--steps", type=int, default=4, help="Inference steps (default: 4)")
+    parser.add_argument("--image", type=str, default=None, help="Path to input image for image-to-image generation")
+    parser.add_argument("--strength", type=float, default=0.75, help="Strength of transformation (0.0-1.0), default 0.75. Higher = more change.")
     args = parser.parse_args()
     
-    generate_image(args.prompt, width=args.width, height=args.height, steps=args.steps)
+    generate_image(args.prompt, width=args.width, height=args.height, steps=args.steps, image_path=args.image, strength=args.strength)