Changes to be committed:

app.py

@@ -1,10 +1,57 @@
 import gradio as gr
 from gradio_client import Client, handle_file
+from PIL import Image, ImageOps
+import os
+import tempfile
 
-def image_to_image(prompt, image, flux_client):
-    prompt = "You are an AI image-to-image editor that transforms a user-provided input image based on their prompt while maintaining consistency and fidelity to the original image. Generate an output image that accurately reflects the user's requested modifications, preserving key elements like style, composition, and context from the input image unless explicitly instructed otherwise. " + prompt
+def resize_and_pad(image_path, dimension):
+    # Open the image
+    img = Image.open(image_path)
+    original_width, original_height = img.size
+
+    if dimension == "Square (1024x1024)":
+        target_width, target_height = 1024, 1024
+    elif dimension == "Portrait (768x1360)":
+        target_width, target_height = 768, 1360
+    elif dimension == "Landscape (1360x768)":
+        target_width, target_height = 1360, 768
+    else:
+        raise ValueError("Invalid dimension selected")
+
+    # Calculate the scaling factor
+    width_ratio = target_width / original_width
+    height_ratio = target_height / original_height
+    scaling_factor = min(width_ratio, height_ratio)
+
+    # Calculate new dimensions
+    new_width = int(original_width * scaling_factor)
+    new_height = int(original_height * scaling_factor)
+
+    # Resize the image
+    img = img.resize((new_width, new_height), Image.LANCZOS)
+
+    # Create a new white image with target dimensions
+    new_img = Image.new("RGB", (target_width, target_height), (255, 255, 255))
+
+    # Calculate padding
+    offset = ((target_width - new_width) // 2, (target_height - new_height) // 2)
+
+    # Paste the resized image into the new image
+    new_img.paste(img, offset)
+
+    # Save the new image temporarily
+    padded_image_path = os.path.join(tempfile.gettempdir(), "padded_image.png")
+    new_img.save(padded_image_path)
+
+    return padded_image_path
+
+def image_to_image(prompt, image, dimension, flux_client):
+    # First, resize and pad the image
+    padded_image_path = resize_and_pad(image, dimension)
+
+    prompt = "Transforms a user-provided input image based on their prompt while maintaining consistency and fidelity to the original image. Generate an output image that accurately reflects the user's requested modifications. " + prompt
     image = flux_client.predict(
-        input_image=handle_file(image),
+        input_image=handle_file(padded_image_path),
         prompt=prompt,
         seed=0,
         randomize_seed=True,
@@ -13,25 +60,30 @@ def image_to_image(prompt, image, flux_client):
         api_name="/infer"
     )
     print(image)
+    # Clean up the temporary file (optional, but good practice)
+    try:
+        os.remove(padded_image_path)
+    except:
+        pass
     return image[0]
 
 def image_to_video(prompt, image, ltx_client):
     result = ltx_client.predict(
-            prompt=prompt,
-            input_image_url=None,
-            middle_image_url=handle_file(image),
-            final_image_url=None,
-            duration_ui=4,
-            api_name="/generate_video"
+        prompt=prompt,
+        input_image_url=None,
+        middle_image_url=handle_file(image),
+        final_image_url=None,
+        duration_ui=4,
+        api_name="/generate_video"
     )
     print(result)
     return result
 
-def personalized_video(prompt, image, request: gr.Request):
+def personalized_video(prompt, image, dimension, request: gr.Request):
     x_ip_token = request.headers['x-ip-token']
     flux_client = Client("black-forest-labs/FLUX.1-Kontext-Dev", headers={"x-ip-token": x_ip_token})
     ltx_client = Client("KingNish/ltx-video-distilled", headers={"x-ip-token": x_ip_token})
-    image = image_to_image(prompt, image, flux_client)
+    image = image_to_image(prompt, image, dimension, flux_client)
     yield image, None
     video = image_to_video(prompt, image, ltx_client)
     yield image, video
@@ -42,22 +94,12 @@ with gr.Blocks() as demo:
         with gr.Column():
             input_image = gr.Image(label="Input Image", type="filepath")
             prompt = gr.Textbox(label="Prompt")
+            dimension = gr.Radio(["Square (1024x1024)", "Portrait (768x1360)", "Landscape (1360x768)"], label="Dimension", value="Square (1024x1024)")
             submit_button = gr.Button("Submit")
         with gr.Column():
             edited_image = gr.Image(label="Edited Image")
             output_video = gr.Video(label="Output Video")
-    submit_button.click(personalized_video, [prompt, input_image], [edited_image, output_video])
-    prompt.submit(personalized_video, [prompt, input_image], [edited_image, output_video])
-
-    # gr.Examples(
-    #     examples=[
-    #         ["A sunny day", "https://black-forest-labs-flux-1-kontext-dev.hf.space/gradio_api/file=/tmp/gradio/927b05b08b8ec2201f7d95722471d2089d4197f152c9c07d0bc9f9672962d03b/flowers.png", None, None],
-    #     ],
-    #     inputs=[prompt, input_image, flux_client, ltx_client],
-    #     outputs=output_video,
-    #     fn=personalized_video,
-    #     cache_examples="lazy"
-    # )
+    submit_button.click(personalized_video, [prompt, input_image, dimension], [edited_image, output_video])
+    prompt.submit(personalized_video, [prompt, input_image, dimension], [edited_image, output_video])
 
 demo.launch()
-