Update app.py

app.py

@@ -6,6 +6,7 @@ from io import BytesIO
 import base64
 import os
 import json
+import random
 
 # Initialize the Google Generative AI client with the API key from environment variables
 try:
@@ -14,38 +15,61 @@ except KeyError:
     raise ValueError("Please set the GEMINI_API_KEY environment variable.")
 client = genai.Client(api_key=api_key)
 
-def generate_item(tag):
+def generate_item(tag, item_index):
     """
-    Generate a single feed item consisting of text from Gemini LLM and an image from Imagen.
+    Generate a single feed item with diverse text and image.
     
     Args:
         tag (str): The tag to base the content on.
+        item_index (int): Index of the item to ensure diversity.
     
     Returns:
         dict: A dictionary with 'text' (str) and 'image_base64' (str).
     """
-    # Generate text using Gemini LLM with JSON output
+    # Define varied styles for diversity in image generation
+    styles = [
+        "futuristic neon lighting",
+        "soft pastel tones with a dreamy vibe",
+        "vibrant and colorful pop art style",
+        "minimalist black and white aesthetic",
+        "retro 80s synthwave look",
+        "golden hour sunlight with warm tones"
+    ]
+    perspectives = [
+        "a close-up view",
+        "a wide-angle shot",
+        "an aerial perspective",
+        "a side profile",
+        "a dynamic angled shot"
+    ]
+    style = random.choice(styles)
+    perspective = random.choice(perspectives)
+
+    # Generate text with high temperature for diversity
     prompt = f"""
     Generate a short, engaging TikTok-style caption about {tag}. 
     Return the response as a JSON object with a single key 'caption' containing the caption text.
     Example: {{"caption": "Craving this yummy treat! 😍 #foodie"}}
     Do not include additional commentary or options.
+    Use creative and varied language to ensure uniqueness.
     """
     text_response = client.models.generate_content(
         model='gemini-2.5-flash-preview-04-17',
-        contents=[prompt]
+        contents=[prompt],
+        generation_config={"temperature": 1.2}  # High temperature for diversity
     )
     # Parse JSON response to extract the caption
     try:
         response_json = json.loads(text_response.text.strip())
         text = response_json['caption']
     except (json.JSONDecodeError, KeyError):
-        text = f"Wow, {tag} is amazing! 😍 #{tag}"  # Fallback caption
+        text = f"Obsessed with {tag}! 🔥 #{tag}"  # Fallback caption
 
-    # Generate an image based on the tag, avoiding text
+    # Generate a diverse image based on the tag
     image_prompt = f"""
-    A vivid, high-quality visual scene representing {tag}, designed for a TikTok video. 
-    The image should be colorful and engaging, with no text or letters included.
+    A high-quality visual scene representing {tag}, designed for a TikTok video. 
+    The image should be {perspective} with a {style}. 
+    Ensure the image is colorful, engaging, and has no text or letters.
     """
     image_response = client.models.generate_images(
         model='imagen-3.0-generate-002',
@@ -63,7 +87,7 @@ def generate_item(tag):
         image = Image.open(BytesIO(generated_image.image.image_bytes))
     else:
         # Fallback to a placeholder image
-        image = Image.new('RGB', (300, 533), color='gray')  # 9:16 aspect ratio
+        image = Image.new('RGB', (360, 640), color='gray')  # 9:16 aspect ratio
 
     # Convert the image to base64
     buffered = BytesIO()
@@ -83,15 +107,15 @@ def start_feed(tag):
         tuple: (current_tag, feed_items, html_content)
     """
     if not tag.strip():
-        tag = "trending"  # Default tag if empty
-    item = generate_item(tag)
+        tag = "trending"
+    item = generate_item(tag, 0)
     feed_items = [item]
     html_content = generate_html(feed_items)
     return tag, feed_items, html_content
 
 def load_more(current_tag, feed_items):
     """
-    Append a new item to the existing feed using the current tag.
+    Append a new item to the existing feed and scroll to the latest item.
     
     Args:
         current_tag (str): The tag currently being used for the feed.
@@ -100,23 +124,24 @@ def load_more(current_tag, feed_items):
     Returns:
         tuple: (current_tag, updated_feed_items, updated_html_content)
     """
-    new_item = generate_item(current_tag)
+    new_item = generate_item(current_tag, len(feed_items))
     feed_items.append(new_item)
-    html_content = generate_html(feed_items)
+    html_content = generate_html(feed_items, scroll_to_latest=True)
     return current_tag, feed_items, html_content
 
-def generate_html(feed_items):
+def generate_html(feed_items, scroll_to_latest=False):
     """
-    Generate an HTML string to display the feed items in a TikTok-like vertical layout.
+    Generate an HTML string to display the feed items in a TikTok-like carousel.
     
     Args:
         feed_items (list): List of dictionaries containing 'text' and 'image_base64'.
+        scroll_to_latest (bool): Whether to auto-scroll to the latest item.
     
     Returns:
         str: HTML string representing the feed.
     """
     html_str = """
-    <div style="
+    <div id="feed-container" style="
         display: flex;
         flex-direction: column;
         align-items: center;
@@ -124,24 +149,28 @@ def generate_html(feed_items):
         margin: 0 auto;
         background-color: #000;
         height: 640px;
-        overflow-y: auto;
+        overflow-y: scroll;
+        scroll-snap-type: y mandatory;
         scrollbar-width: none;
         -ms-overflow-style: none;
         border: 1px solid #333;
         border-radius: 10px;
     ">
     """
-    # Hide scrollbar for a cleaner look
+    # Hide scrollbar
     html_str += """
     <style>
-        div::-webkit-scrollbar {
+        #feed-container::-webkit-scrollbar {
             display: none;
         }
+        .feed-item {
+            scroll-snap-align: start;
+        }
     </style>
     """
-    for item in feed_items:
+    for idx, item in enumerate(feed_items):
         html_str += f"""
-        <div style="
+        <div class="feed-item" id="item-{idx}" style="
             width: 100%;
             height: 640px;
             position: relative;
@@ -175,6 +204,15 @@ def generate_html(feed_items):
         </div>
         """
     html_str += "</div>"
+
+    # Auto-scroll to the latest item if requested
+    if scroll_to_latest and feed_items:
+        html_str += f"""
+        <script>
+            document.getElementById('item-{len(feed_items) - 1}').scrollIntoView({{ behavior: 'smooth' }});
+        </script>
+        """
+
     return html_str
 
 # Define the Gradio interface