Update app.py

app.py

@@ -1,42 +1,18 @@
-import streamlit as st
-from PIL import Image
+import gradio as gr
 import torch
-import numpy as np
-
-# Import the Diffusers pipeline and a utility for LoRA (adjust as needed)
+from PIL import Image
 from diffusers import StableDiffusionImg2ImgPipeline
 
-# Optional: If you have a helper to apply LoRA weights, import it here.
-# For example, if you use lora_diffusion, you might do:
-# from lora_diffusion import monkeypatch_lora, tune_lora_scale
-
-st.set_page_config(page_title="GHIBSKY Art Transformer", layout="centered")
-
-st.title("GHIBSKY Art Transformer")
-st.write(
-    """
-    Upload your portrait image and see it transformed into enchanting, Ghibli-inspired art.
-    
-    This demo applies the Flux Ghibsky Illustration model locally. The model blends serene, surreal skies 
-    with intricate, Ghibli-inspired details. To invoke the unique aesthetic, the prompt begins with **GHIBSKY style**.
-    
-    *Note: This demo caches files only within the session; no data is stored permanently.*
-    """
-)
-
-@st.cache_resource
 def load_model():
     """
     Load the base Stable Diffusion model and apply the LoRA weights locally.
-    Replace the placeholder code with your actual LoRA integration as needed.
+    Replace the placeholder code with your actual LoRA integration.
     """
-    model_id = "runwayml/stable-diffusion-v1-5"  # Change if using a different base model.
+    model_id = "runwayml/stable-diffusion-v1-5"  # Change to your base model if needed.
     pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
         model_id, 
         torch_dtype=torch.float16
     )
-    
-    # Send model to GPU if available.
     device = "cuda" if torch.cuda.is_available() else "cpu"
     pipe = pipe.to(device)
     
@@ -44,36 +20,47 @@ def load_model():
     lora_path = "./flux_ghibsky_lora.safetensors"
     
     # Apply the LoRA weights to the model.
-    # The following is a placeholder. Use your actual LoRA integration.
-    # Example using a hypothetical monkeypatch_lora:
+    # Replace the following placeholder with your actual integration code.
+    # For example, if using a helper like monkeypatch_lora:
     # monkeypatch_lora(pipe.unet, lora_path)
     # tune_lora_scale(pipe.unet, 1.0)
     
-    st.write("Model loaded and LoRA weights applied successfully.")
+    print("Model loaded and LoRA weights applied.")
     return pipe
 
+# Load the model once at startup.
 pipe = load_model()
 
-uploaded_file = st.file_uploader("Upload your portrait image", type=["jpg", "jpeg", "png"])
+def transform_image(image: Image.Image, strength: float, steps: int) -> Image.Image:
+    """
+    Transforms the uploaded image into Ghibli-inspired art.
+    The prompt starts with the trigger word "GHIBSKY style".
+    """
+    prompt = (
+        "GHIBSKY style, a portrait transformed into dreamy, Ghibli-inspired art, "
+        "featuring serene skies, surreal details, and intricate brush strokes"
+    )
+    
+    result = pipe(prompt=prompt, image=image, strength=strength, num_inference_steps=steps)
+    return result.images[0]
 
-if uploaded_file:
-    # Open and display the original image.
-    original_image = Image.open(uploaded_file).convert("RGB")
-    st.image(original_image, caption="Original Image", use_column_width=True)
+# Create a Gradio interface.
+demo = gr.Interface(
+    fn=transform_image,
+    inputs=[
+        gr.Image(type="pil", label="Upload your portrait image"),
+        gr.Slider(0.1, 0.9, value=0.6, label="Transformation Strength"),
+        gr.Slider(20, 100, step=5, value=50, label="Inference Steps")
+    ],
+    outputs=gr.Image(type="pil", label="Ghibli-Inspired Art"),
+    title="GHIBSKY Art Transformer",
+    description=(
+        "Upload your portrait image and see it transformed into enchanting, Ghibli-inspired art. "
+        "This demo uses the Flux Ghibsky Illustration model locally. The prompt is automatically "
+        "prefixed with 'GHIBSKY style' to invoke the model's unique aesthetic. "
+        "No data is stored permanently."
+    )
+)
 
-    # Define the prompt with the trigger word.
-    prompt = ("GHIBSKY style, a portrait transformed into dreamy, Ghibli-inspired art, "
-              "featuring serene skies, surreal details, and intricate brush strokes")
-    
-    # Optional: Adjust strength and number of inference steps as needed.
-    strength = st.slider("Transformation Strength", 0.1, 0.9, 0.6, step=0.05)
-    num_inference_steps = st.slider("Inference Steps", 20, 100, 50, step=5)
-    
-    st.write("Transforming your image...")
-    
-    # Run image-to-image transformation.
-    # Note: The image is resized internally by the pipeline if needed.
-    result = pipe(prompt=prompt, image=original_image, strength=strength, num_inference_steps=num_inference_steps)
-    
-    transformed_image = result.images[0]
-    st.image(transformed_image, caption="Ghibli-Inspired Art", use_column_width=True)
+if __name__ == "__main__":
+    demo.launch()