Update app.py

app.py

@@ -11,6 +11,9 @@ from pipeline_fill_sd_xl import StableDiffusionXLFillPipeline
 
 from PIL import Image, ImageDraw
 import numpy as np
+import cv2
+import tempfile
+import os
 
 config_file = hf_hub_download(
     "xinsir/controlnet-union-sdxl-1.0",
@@ -43,7 +46,6 @@ pipe = StableDiffusionXLFillPipeline.from_pretrained(
 
 pipe.scheduler = TCDScheduler.from_config(pipe.scheduler.config)
 
-
 def can_expand(source_width, source_height, target_width, target_height, alignment):
     """Checks if the image can be expanded based on the alignment."""
     if alignment in ("Left", "Right") and source_width >= target_width:
@@ -89,6 +91,7 @@ def infer(image, width, height, overlap_width, num_inference_steps, resize_optio
 
     if not can_expand(source.width, source.height, target_size[0], target_size[1], alignment):
         alignment = "Middle"
+
     # Calculate margins based on alignment
     if alignment == "Middle":
         margin_x = (target_size[0] - source.width) // 2
@@ -200,6 +203,41 @@ def select_the_right_preset(user_width, user_height):
 def toggle_custom_resize_slider(resize_option):
     return gr.update(visible=(resize_option == "Custom"))
 
+def create_video_from_images(image_list, fps=8):
+    if not image_list:
+        return None
+    
+    with tempfile.NamedTemporaryFile(delete=False, suffix='.mp4') as temp_video_file:
+        video_path = temp_video_file.name
+
+    frame = np.array(image_list[0])
+    height, width, layers = frame.shape
+    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
+    video = cv2.VideoWriter(video_path, fourcc, fps, (width, height))
+
+    for image in image_list:
+        video.write(cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR))
+
+    video.release()
+    return video_path
+
+def loop_outpainting(image, width, height, overlap_width, num_inference_steps, resize_option, custom_resize_size, prompt_input, alignment, num_iterations):
+    image_list = []
+    current_image = image
+
+    for _ in range(num_iterations):
+        # Generate new image
+        result = next(infer(current_image, width, height, overlap_width, num_inference_steps, resize_option, custom_resize_size, prompt_input, alignment))
+        new_image = result[1]
+        image_list.append(new_image)
+        
+        # Use new image as input for next iteration
+        current_image = new_image
+
+    # Create video from image list
+    video_path = create_video_from_images(image_list)
+    return video_path
+
 css = """
 .gradio-container {
     width: 1200px !important;
@@ -288,6 +326,8 @@ with gr.Blocks(css=css) as demo:
                                 value=512,
                                 visible=False
                             )
+                        with gr.Row():
+                            num_iterations = gr.Slider(label="Number of iterations", minimum=2, maximum=1024, step=1, value=10)
                             
                 gr.Examples(
                     examples=[
@@ -305,8 +345,11 @@ with gr.Blocks(css=css) as demo:
                     label="Generated Image",
                 )
                 use_as_input_button = gr.Button("Use as Input Image", visible=False)
+                loop_button = gr.Button("Loop Outpainting")
+                video_output = gr.Video(label="Outpainting Video")
 
     def use_output_as_input(output_image):
+        """Sets the generated output as the new input image."""
         return gr.update(value=output_image[1])
 
     use_as_input_button.click(
@@ -373,4 +416,11 @@ with gr.Blocks(css=css) as demo:
         outputs=use_as_input_button,
     )
 
+    loop_button.click(
+        fn=loop_outpainting,
+        inputs=[input_image, width_slider, height_slider, overlap_width, num_inference_steps,
+                resize_option, custom_resize_size, prompt_input, alignment_dropdown, num_iterations],
+        outputs=video_output,
+    )
+
 demo.queue(max_size=12).launch(share=False)