smart crop

app.py

@@ -239,24 +239,74 @@ def find_place_to_add_text(image: Image.Image, faces: list[dict]) -> tuple[int,
         min_text_width,  # width
         min_text_height  # height
     )
+    
+def crop_to_ratio_while_preventing_faces(image: Image.Image, faces: list[dict]) -> Image.Image:
+    ASPECT_RATIO_PORTRAIT = 5/7
+    ASPECT_RATIO_LANDSCAPE = 7/5
+    image_width, image_height = image.size
+    
+    # Calculate current aspect ratio
+    current_ratio = image_width / image_height
+    is_portrait = current_ratio < 1
+    target_ratio = ASPECT_RATIO_PORTRAIT if is_portrait else ASPECT_RATIO_LANDSCAPE
+    
+    # Calculate new dimensions
+    if current_ratio > target_ratio:
+        new_width = int(image_height * target_ratio)
+        new_height = image_height
+    else:
+        new_width = image_width
+        new_height = int(image_width / target_ratio)
+    
+    # If no faces, just do center crop
+    if not faces:
+        x = (image_width - new_width) // 2
+        y = (image_height - new_height) // 2
+        return image.crop((x, y, x + new_width, y + new_height))
+    
+    # Find the bounding box that contains all faces
+    face_x1 = min(int(face['bbox'][0]) for face in faces)
+    face_y1 = min(int(face['bbox'][1]) for face in faces)
+    face_x2 = max(int(face['bbox'][2]) for face in faces)
+    face_y2 = max(int(face['bbox'][3]) for face in faces)
+    
+    # Add padding around faces
+    padding = 50
+    face_x1 = max(0, face_x1 - padding)
+    face_y1 = max(0, face_y1 - padding)
+    face_x2 = min(image_width, face_x2 + padding)
+    face_y2 = min(image_height, face_y2 + padding)
+    
+    # Calculate crop coordinates that ensure faces are included
+    x = max(0, min(face_x1, image_width - new_width))
+    y = max(0, min(face_y1, image_height - new_height))
+    
+    # Adjust if faces would be cut off
+    if x + new_width < face_x2:
+        x = max(0, face_x2 - new_width)
+    if y + new_height < face_y2:
+        y = max(0, face_y2 - new_height)
+    
+    return image.crop((x, y, x + new_width, y + new_height))
 
 def run_flow(input_image, holiday, message):
     faces = detect_faces(input_image)
-    
-    prompt = image_to_prompt(input_image, holiday)
+    cropped_image = crop_to_ratio_while_preventing_faces(input_image, faces)
+            
+    prompt = image_to_prompt(cropped_image, holiday)
     print(prompt)
-    result_image, only_background_image, mask = remove_background(input_image)
+    result_image, only_background_image, mask = remove_background(cropped_image)
     dilated_mask = dilate_mask(mask)
-    output_image = modify_background(input_image, dilated_mask, prompt)
+    output_image = modify_background(cropped_image, dilated_mask, prompt)
     
     # Create a copy of the modified image before drawing
     output_image_with_text_rectangle = output_image.copy()
-    text_x_in_percent, text_y_in_percent, text_width_in_percent, text_height_in_percent = find_place_to_add_text(input_image, faces)
+    text_x_in_percent, text_y_in_percent, text_width_in_percent, text_height_in_percent = find_place_to_add_text(cropped_image, faces)
     text_x = text_x_in_percent * output_image.width
     text_y = text_y_in_percent * output_image.height
     text_width = text_width_in_percent * output_image.width
     text_height = text_height_in_percent * output_image.height
-    print(text_x, text_y, text_width, text_height)
+
     draw = ImageDraw.Draw(output_image_with_text_rectangle)
     draw.rectangle((text_x, text_y, text_x + text_width, text_y + text_height), outline="red")