Update app.py

app.py

@@ -2,15 +2,15 @@ import os
 import sys
 import logging
 
-# Logging setup
+# Logging configuration
 logging.basicConfig(level=logging.INFO, 
                     format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
                     handlers=[logging.StreamHandler(sys.stdout)])
 logger = logging.getLogger(__name__)
 
-# Auto-install dependencies
+# Install required dependencies automatically
 def install_dependencies():
-    logger.info("Checking and installing required dependencies...")
+    logger.info("Verifying and installing required dependencies...")
     try:
         # Try to import peft
         try:
@@ -29,15 +29,15 @@ def install_dependencies():
             os.system("pip install -q bitsandbytes>=0.41.0")
             
         # Ensure other dependencies are installed
-        logger.info("Checking other dependencies...")
+        logger.info("Verifying other dependencies...")
         os.system("pip install -q torch transformers>=4.30.0 accelerate>=0.20.0 gradio pillow psutil")
         
-        logger.info("All dependencies installed successfully")
+        logger.info("All dependencies successfully installed")
         
         # Re-import peft to verify
         import peft
         from peft import PeftModel, PeftConfig
-        logger.info(f"PEFT imported successfully, version: {peft.__version__}")
+        logger.info(f"PEFT correctly imported, version: {peft.__version__}")
         
         return True
     except Exception as e:
@@ -47,9 +47,9 @@ def install_dependencies():
 # Install dependencies before importing
 success = install_dependencies()
 if not success:
-    logger.error("Failed to install required dependencies. The application may not work properly.")
+    logger.error("Failed to install required dependencies. The application may not function properly.")
 
-# Now that we have dependencies, import modules
+# Now that we have the dependencies, we import the modules
 import torch
 from transformers import BlipProcessor, BlipForConditionalGeneration, AutoModelForCausalLM, AutoTokenizer
 from peft import PeftModel, PeftConfig
@@ -66,15 +66,15 @@ if use_gpu:
         logger.info(f"Total GPU memory: {torch.cuda.get_device_properties(0).total_memory / 1024**3:.2f} GB")
         logger.info(f"Available GPU memory: {torch.cuda.memory_reserved(0) / 1024**3:.2f} GB")
     except:
-        logger.info("Could not get detailed GPU information")
+        logger.info("Could not retrieve detailed GPU information")
 
 # Lazy loading of models
 processor, model = None, None
 peft_model, tokenizer = None, None
 
-# Custom function to generate text with PEFT model
+# Custom function to generate text with the PEFT model
 def generate_with_peft_model(prompt, max_new_tokens=100, temperature=0.7, top_p=0.95):
-    """Generate text directly using PEFT model without pipeline"""
+    """Generates text using the PEFT model directly without pipeline"""
     global peft_model, tokenizer
     
     if peft_model is None or tokenizer is None:
@@ -105,7 +105,7 @@ def generate_with_peft_model(prompt, max_new_tokens=100, temperature=0.7, top_p=
             response = output_text.split("<|assistant|>")[-1].strip()
             return response
         
-        # If we can't extract assistant response, remove original prompt
+        # If we can't extract assistant response, remove the original prompt
         if prompt in output_text:
             response = output_text[len(prompt):].strip()
             return response
@@ -198,13 +198,13 @@ def load_models():
                     base_model,
                     local_adapter_path
                 )
-                logger.info("✅ LORA adapter loaded successfully from local path")
+                logger.info("✅ LORA adapter loaded successfully from local storage")
                 
                 return True
                 
             except Exception as e2:
                 logger.error(f"Error loading LORA adapter locally: {str(e2)}")
-                logger.error("Could not load LORA adapter. The application may not work properly.")
+                logger.error("Could not load LORA adapter. The application will not function properly.")
                 return False
                 
     except Exception as e:
@@ -213,29 +213,17 @@ def load_models():
         logger.error(traceback.format_exc())
         return False
 
-# Universal Video Prompting Guide combining Gen-4 + SARA
+# Universal Video Prompting Guide combining SARA framework
 unified_instructions = """
 # 🎬 Universal Video Prompting Guide
-*Compatible with Gen-4, Sora, Pika, Luma, Runway and all diffusion-based video models*
-## Core Principles (Universal)
+*Compatible with Sora, Gen-4, Pika, Luma, Runway and all diffusion-based video models*
+## Core Principles
 ✅ **Focus on MOTION, not static description**
 ✅ **Use positive phrasing exclusively**
 ✅ **Start simple, iterate progressively**
 ✅ **Refer to subjects in general terms** ("the subject," "the woman")
 ✅ **Keep prompts direct and easily understood**
-## Two Complementary Approaches
-### 🚀 **Gen-4 Official Method** (Recommended for beginners)
-**Structure**: Simple iterative building
-1. Start with essential motion only
-2. Add one element at a time: Subject Motion → Camera Motion → Scene Motion → Style Descriptors
-3. Use general terms and avoid complex descriptions
-**Example**:
-- Basic: "The subject walks forward"
-- + Camera: "The subject walks forward. Handheld camera follows"
-- + Scene: "The subject walks forward. Handheld camera follows. Dust trails behind"
-- + Style: "The subject walks forward. Handheld camera follows. Dust trails behind. Cinematic."
-### 🎯 **SARA Framework** (Advanced precision)
-**Structure**: [Subject] + [Action] + [Reference] + [Atmosphere]
+## SARA Framework (Subject + Action + Reference + Atmosphere)
 - **Subject (S)**: Main element to control
 - **Action (A)**: Movement/transformation ([verb] + [adverb])
 - **Reference (R)**: Spatial anchors ("while X remains steady")
@@ -324,7 +312,7 @@ def analyze_scene_with_zephyr(basic_caption, aspect_ratio, composition):
     """Use PEFT model for advanced scene analysis"""
     logger.info("Starting scene analysis...")
     
-    # Verify model is loaded
+    # Verify that the model is loaded
     if peft_model is None or tokenizer is None:
         logger.error("PEFT model not available")
         return {
@@ -344,8 +332,8 @@ Please provide:
 1. Type of motion that would work best
 2. Recommended camera movements
 3. Emotional tone/style suggestions
-4. Best prompting approach (SARA vs Gen-4)
-Be concise and practical. Keep your response in English.
+4. Best prompting approach (SARA framework)
+Be concise and practical.
 <|assistant|>"""
 
         logger.info("Generating analysis with PEFT model...")
@@ -356,7 +344,7 @@ Be concise and practical. Keep your response in English.
             top_p=0.95
         )
         
-        logger.info(f"Generated analysis: {generated_text[:100]}...")
+        logger.info(f"Analysis generated: {generated_text[:100]}...")
         
         lines = generated_text.split('\n')
         motion_insights = []
@@ -382,7 +370,7 @@ Be concise and practical. Keep your response in English.
         import traceback
         logger.error(traceback.format_exc())
         return {
-            'scene_interpretation': f"Error in analysis: {str(e)}",
+            'scene_interpretation': f"Analysis error: {str(e)}",
             'motion_insights': ["Error during analysis", "Try with another image"],
             'recommended_approach': "SARA framework (default)"
         }
@@ -391,7 +379,7 @@ def generate_sample_prompts_with_zephyr(scene_info=None):
     """Generate sample prompts using PEFT model"""
     logger.info("Generating sample prompts...")
     
-    # Verify model is loaded
+    # Verify that the model is loaded
     if peft_model is None or tokenizer is None:
         logger.error("PEFT model not available")
         return [
@@ -404,14 +392,12 @@ def generate_sample_prompts_with_zephyr(scene_info=None):
         try:
             # Use PEFT model to generate contextual prompts
             context_prompt = f"""<|system|>
-Generate 3 professional video prompts using the SARA framework based on this image analysis. 
-Each prompt should follow the structure: Subject + Action + Reference + Atmosphere.
-Ensure the prompts are in English, emphasize motion, and are compatible with AI video models.
+Generate 3 professional video prompts using the SARA framework based on this image analysis.
 <|user|>
 Image description: {scene_info['basic_description']}
 Composition: {scene_info.get('composition', 'Balanced')}
 Aspect Ratio: {scene_info.get('aspect_ratio', 'N/A'):.2f}
-Create three unique and expressive prompts following the SARA framework.
+Remember the SARA framework: Subject + Action + Reference + Atmosphere
 <|assistant|>"""
 
             logger.info("Generating prompts for the scene...")
@@ -453,7 +439,7 @@ def optimize_user_prompt_with_zephyr(user_idea, scene_info=None):
     if not user_idea.strip():
         return "Please enter your idea first.", "No input provided"
 
-    # Verify model is loaded
+    # Verify that the model is loaded
     if peft_model is None or tokenizer is None:
         logger.error("PEFT model not available")
         return "Error: Model not available. Try reloading the application.", "Model not loaded"
@@ -467,23 +453,41 @@ def optimize_user_prompt_with_zephyr(user_idea, scene_info=None):
     try:
         # Enforce structure based on approach
         logger.info("Preparing prompt for optimization...")
-        optimization_prompt = f"""<|system|>
-You are an expert in video prompting, specializing in the SARA framework for advanced video prompt creation.
-Your task is to transform user ideas into professional, optimized prompts for AI video models.
-
+        
+        # Detect language and adjust system prompt accordingly
+        import re
+        non_english_pattern = re.compile(r'[^\x00-\x7F]+')
+        has_non_english = bool(non_english_pattern.search(user_idea))
+        
+        if has_non_english:
+            logger.info("Detected non-English input")
+            optimization_prompt = f"""<|system|>
+You are an expert in video prompting, specializing in the SARA framework. Transform user ideas into professional prompts compatible with AI video models like Sora, Gen-4, Pika, Runway, and Luma.
+IMPORTANT: Preserve the original language of the user's idea in your response. For example, if they write in Spanish, your response should be in Spanish.
 Key principles:
 - Focus on MOTION, not static description
-- Use positive phrasing only
+- Use positive phrasing
 - Be specific about camera work
 - Include lighting/atmosphere details
-- Always follow the SARA structure: Subject + Action + Reference + Atmosphere
-- Always write prompts in English, regardless of the language of the input
-
-For any input in any language, create an optimized English video prompt following the SARA framework.
+- Follow the SARA structure: Subject + Action + Reference + Atmosphere
+<|user|>
+User's idea: "{user_idea}"
+{context}
+Please create an optimized video prompt using the SARA framework. Respond with just the prompt in the same language as the user's input.
+<|assistant|>"""
+        else:
+            optimization_prompt = f"""<|system|>
+You are an expert in video prompting, specializing in the SARA framework. Transform user ideas into professional prompts compatible with AI video models like Sora, Gen-4, Pika, Runway, and Luma.
+Key principles:
+- Focus on MOTION, not static description
+- Use positive phrasing
+- Be specific about camera work
+- Include lighting/atmosphere details
+- Follow the SARA structure: Subject + Action + Reference + Atmosphere
 <|user|>
 User's idea: "{user_idea}"
 {context}
-Create a professional video prompt using the SARA framework. Respond with just the prompt in English.
+Please create an optimized video prompt using the SARA framework. Respond with just the prompt.
 <|assistant|>"""
 
         logger.info("Generating optimized prompt...")
@@ -495,6 +499,8 @@ Create a professional video prompt using the SARA framework. Respond with just t
         )
         
         logger.info(f"Optimized prompt: {optimized}")
+        
+        # Status message in English regardless of input language
         return optimized, "SARA-Zephyr LORA used successfully"
         
     except Exception as e:
@@ -505,7 +511,7 @@ Create a professional video prompt using the SARA framework. Respond with just t
                 f"Error: {str(e)}")
 
 def fallback_generate_prompt(user_idea, scene_info=None):
-    """Fallback function to generate prompts manually if model fails"""
+    """Fallback function to generate prompts manually if the model fails"""
     logger.info(f"Using fallback generation for: {user_idea}")
     
     if not user_idea.strip():
@@ -539,7 +545,7 @@ def refine_prompt_with_zephyr(current_prompt, feedback, chat_history, scene_info
     if not feedback.strip():
         return current_prompt, chat_history
         
-    # Verify model is loaded
+    # Verify that the model is loaded
     if peft_model is None or tokenizer is None:
         logger.error("PEFT model not available")
         return "Error: Model not available. Try reloading the application.", chat_history
@@ -550,26 +556,32 @@ def refine_prompt_with_zephyr(current_prompt, feedback, chat_history, scene_info
         context = f"Image context: {scene_info['basic_description']}"
 
     try:
+        # Detect language of current prompt and feedback
+        import re
+        non_english_pattern = re.compile(r'[^\x00-\x7F]+')
+        has_non_english_prompt = bool(non_english_pattern.search(current_prompt))
+        has_non_english_feedback = bool(non_english_pattern.search(feedback))
+        
+        # Determine response language
+        preserve_language_instruction = ""
+        if has_non_english_prompt or has_non_english_feedback:
+            preserve_language_instruction = "IMPORTANT: Preserve the original language of the prompt in your response. For example, if the prompt is in Spanish, your refined prompt should be in Spanish."
+        
         # Construct refinement prompt
         refinement_prompt = f"""<|system|>
-You are an expert in refining video prompts using the SARA framework. You are given an existing prompt and specific feedback on how to improve it.
-Your task is to create an enhanced version of the prompt that incorporates the feedback while maintaining the SARA structure.
-
+You are an expert in refining video prompts using the SARA framework. Based on the user's feedback, improve the current prompt while maintaining its core structure.
+{preserve_language_instruction}
 Key principles:
 - Focus on MOTION, not static description
-- Use positive phrasing only
+- Use positive phrasing
 - Be specific about camera work
 - Include lighting/atmosphere details
-- Always follow the SARA structure: Subject + Action + Reference + Atmosphere
-- Always keep the prompt in English
-- Apply the requested changes precisely as mentioned in the feedback
-
-Produce only the refined prompt text, nothing else.
+- Follow the SARA structure: Subject + Action + Reference + Atmosphere
 <|user|>
 Current prompt: "{current_prompt}"
 Feedback: "{feedback}"
 {context}
-Please refine the prompt based on this feedback. Keep it under 100 words. Return only the refined prompt.
+Please refine the prompt while keeping it under 100 words. Respond with just the refined prompt.
 <|assistant|>"""
 
         logger.info("Generating refined prompt...")
@@ -592,49 +604,33 @@ Please refine the prompt based on this feedback. Keep it under 100 words. Return
         logger.error(traceback.format_exc())
         return f"Error refining prompt: {str(e)}. Try with a simpler request.", chat_history
 
-def build_custom_prompt(foundation, subject_motion, scene_motion, camera_motion, style, approach="SARA"):
-    """Build custom prompt using selected approach"""
-    if approach == "SARA":
-        # SARA Structure: [Subject] [Action] while [Reference], [Atmosphere]
-        parts = []
-        if foundation:
-            parts.append(foundation)
-
-        # Add motion elements
-        motion_parts = []
-        if subject_motion:
-            motion_parts.extend(subject_motion)
-        if scene_motion:
-            motion_parts.extend(scene_motion)
-        if motion_parts:
-            parts.append(", ".join(motion_parts))
-
-        # Reference (camera stability)
-        if camera_motion:
-            parts.append(f"while {camera_motion}")
-        else:
-            parts.append("while background remains steady")
-
-        # Atmosphere
-        if style:
-            parts.append(style)
-
-        return " ".join(parts)
-    else:  # Gen-4 style
-        # Gen-4 Structure: Simple iterative building
-        parts = []
-        if foundation:
-            parts.append(foundation)
-        if subject_motion:
-            parts.extend(subject_motion)
-        if camera_motion:
-            parts.append(camera_motion)
-        if scene_motion:
-            parts.extend(scene_motion)
-        if style:
-            parts.append(style)
-
-        return ". ".join(parts) if parts else "The subject moves naturally"
+def build_custom_prompt(foundation, subject_motion, scene_motion, camera_motion, style):
+    """Build custom prompt using SARA framework"""
+    # SARA Structure: [Subject] [Action] while [Reference], [Atmosphere]
+    parts = []
+    if foundation:
+        parts.append(foundation)
+
+    # Add motion elements
+    motion_parts = []
+    if subject_motion:
+        motion_parts.extend(subject_motion)
+    if scene_motion:
+        motion_parts.extend(scene_motion)
+    if motion_parts:
+        parts.append(", ".join(motion_parts))
+
+    # Reference (camera stability)
+    if camera_motion:
+        parts.append(f"while {camera_motion}")
+    else:
+        parts.append("while background remains steady")
+
+    # Atmosphere
+    if style:
+        parts.append(style)
+
+    return " ".join(parts)
 
 def test_basic_generation():
     """Test basic generation with PEFT model"""
@@ -687,7 +683,7 @@ def get_debug_info():
         if peft_model is not None:
             info.append(f"PEFT model type: {type(peft_model).__name__}")
             
-            # More info about PEFT model
+            # More information about PEFT model
             if hasattr(peft_model, 'base_model'):
                 base_model_type = type(peft_model.base_model).__name__
                 info.append(f"Base model type: {base_model_type}")
@@ -713,5 +709,291 @@ def get_debug_info():
         info.append(f"GPU available: {torch.cuda.is_available()}")
         if torch.cuda.is_available():
             info.append(f"GPU device: {torch.cuda.get_device_name(0)}")
-            info.append(f"Memory allocated: {torch.cuda.memory_allocated(0) / (1024**3):.2f} GB")
-            info.append(f"Memory reserved: {torch.cuda.memory_reserved(0) / (1024**3):.2f} GB")
+            info.append(f"Allocated memory: {torch.cuda.memory_allocated(0) / (1024**3):.2f} GB")
+            info.append(f"Reserved memory: {torch.cuda.memory_reserved(0) / (1024**3):.2f} GB")
+        
+        # System memory information
+        try:
+            import psutil
+            vm = psutil.virtual_memory()
+            info.append(f"Total RAM: {vm.total / (1024**3):.2f} GB")
+            info.append(f"Available RAM: {vm.available / (1024**3):.2f} GB")
+            info.append(f"RAM usage percentage: {vm.percent}%")
+        except ImportError:
+            info.append("psutil not available for system memory information")
+        
+        return "\n".join(info)
+    except Exception as e:
+        logger.error(f"Error generating debug info: {str(e)}")
+        return f"Error: {str(e)}"
+
+# Create the Gradio interface
+def create_interface():
+    """Create the Gradio interface"""
+    # Pre-load models
+    try:
+        logger.info("Pre-loading models...")
+        load_models()
+    except Exception as e:
+        logger.error(f"Error during preloading: {str(e)}")
+        logger.info("Models will be loaded on demand")
+        
+    logger.info("Creating Gradio interface...")
+    
+    with gr.Blocks(title="AI Video Prompt Generator") as demo:
+        # Header
+        gr.Markdown("# 🎬 AI Video Prompt Generator - 🤖 SARA Framework")
+        gr.Markdown("*Professional prompts for Sora, Gen-4, Pika, Luma, Runway and more*")
+        
+        # State variables
+        scene_state = gr.State({})
+        chat_history_state = gr.State([])
+
+        with gr.Tabs():
+            # Tab 1: Learning Guide
+            with gr.Tab("📚 Prompting Guide"):
+                gr.Markdown(unified_instructions)
+                # Advanced tips
+                with gr.Accordion("🎯 Advanced Tips", open=False):
+                    gr.Markdown("""
+                    ## Advanced Prompting Strategies
+                    ### 🎨 Style Integration
+                    - **Cinematography**: "Dutch angle," "Extreme close-up," "Bird's eye view"
+                    - **Lighting**: "Golden hour," "Neon glow," "Harsh shadows," "Soft diffused light"
+                    - **Movement Quality**: "Fluid motion," "Mechanical precision," "Organic flow"
+                    ### ⚡ Motion Types
+                    - **Subject Motion**: Walking, running, dancing, gesturing
+                    - **Camera Motion**: Pan, tilt, dolly, zoom, orbit, tracking
+                    - **Environmental**: Wind, water flow, particle effects, lighting changes
+                    """)
+
+            # Tab 2: Image Analysis
+            with gr.Tab("📷 Image Analysis"):
+                with gr.Row():
+                    with gr.Column(scale=1):
+                        image_input = gr.Image(
+                            label="Upload Image for Analysis",
+                            type="pil"
+                        )
+                        analyze_btn = gr.Button("🔍 Analyze Image", variant="primary")
+                    with gr.Column(scale=2):
+                        analysis_output = gr.Markdown(label="AI Analysis Results")
+                
+                # Sample prompts section
+                with gr.Group():
+                    gr.Markdown("### 💡 Sample Prompts")
+                    sample_btn = gr.Button("🎲 Generate Sample Prompts")
+                    sample_prompts = [
+                        gr.Textbox(
+                            label=f"Sample {i+1}", 
+                            lines=2, 
+                            interactive=False,
+                            show_copy_button=True
+                        )
+                        for i in range(3)
+                    ]
+
+            # Tab 3: AI Prompt Generator
+            with gr.Tab("🤖 AI Prompt Generator"):
+                with gr.Row():
+                    with gr.Column():
+                        user_idea = gr.Textbox(
+                            label="Your Video Idea (any language)",
+                            placeholder="e.g., 'el personaje camina lentamente' or 'character walks slowly'",
+                            lines=3
+                        )
+                        optimize_btn = gr.Button("🚀 Generate Optimized Prompt", variant="primary")
+                        with gr.Row():
+                            retry_btn = gr.Button("🔄 Manual Generation Fallback", variant="secondary")
+                            model_status = gr.Textbox(
+                                label="Model Status",
+                                value="",
+                                interactive=False
+                            )
+                        optimized_prompt = gr.Textbox(
+                            label="AI-Optimized Video Prompt",
+                            lines=4,
+                            interactive=True,
+                            show_copy_button=True
+                        )
+                        # Basic test button
+                        test_btn = gr.Button("🔬 Test Basic Generation", variant="secondary")
+                        test_output = gr.Textbox(
+                            label="Basic Generation Test",
+                            lines=2,
+                            interactive=False
+                        )
+                    with gr.Column():
+                        gr.Markdown("### 🔄 Refine Your Prompt")
+                        feedback_input = gr.Textbox(
+                            label="Feedback/Changes",
+                            placeholder="e.g., 'make it more dramatic' or 'add camera movement'",
+                            lines=2
+                        )
+                        refine_btn = gr.Button("🔄 Refine Prompt")
+                        # Chat history
+                        with gr.Accordion("💬 Refinement History", open=False):
+                            chat_display = gr.Chatbot(height=300, type='messages')
+                        
+                        # Model status and debug info
+                        with gr.Accordion("🔧 Debug Info", open=False):
+                            debug_info = gr.Textbox(
+                                label="Debug Information",
+                                value="Click 'Get Debug Info' to see model status",
+                                lines=8,
+                                interactive=False
+                            )
+                            debug_btn = gr.Button("Get Debug Info")
+
+            # Tab 4: Custom Builder
+            with gr.Tab("🛠️ Custom Builder"):
+                gr.Markdown("## Build Your Custom Prompt")
+                with gr.Row():
+                    custom_foundation = gr.Textbox(
+                        label="Foundation",
+                        placeholder="The subject...",
+                        lines=1
+                    )
+                with gr.Row():
+                    subject_motion = gr.CheckboxGroup(
+                        choices=[
+                            "walks smoothly", "speaks clearly", "gestures naturally", 
+                            "moves gracefully", "turns slowly", "smiles confidently",
+                            "dances rhythmically", "stands firmly", "runs energetically",
+                            "sits relaxed", "laughs joyfully", "looks curiously"
+                        ],
+                        label="Subject Motion"
+                    )
+                    scene_motion = gr.CheckboxGroup(
+                        choices=[
+                            "dust swirls", "lighting changes", "wind effects", 
+                            "water movement", "atmosphere shifts", "leaves flutter",
+                            "shadows elongate", "fog rolls in", "sunlight filters through",
+                            "rain falls gently", "snow drifts", "crowds bustle"
+                        ],
+                        label="Scene Motion"
+                    )
+                with gr.Row():
+                    camera_motion = gr.Dropdown(
+                        choices=[
+                            "camera remains steady", "handheld camera follows", 
+                            "camera pans left", "camera pans right", 
+                            "camera tracks forward", "camera zooms in slowly",
+                            "camera pulls back", "camera orbits subject",
+                            "drone shot from above", "camera tilts upward",
+                            "camera moves from low angle", "camera shifts focus"
+                        ],
+                        label="Camera Motion",
+                        value="camera remains steady"
+                    )
+                    style_motion = gr.Dropdown(
+                        choices=[
+                            "cinematic atmosphere", "documentary style", "live-action feel",
+                            "dramatic lighting", "peaceful ambiance", "energetic mood",
+                            "professional setting", "nostalgic tone", "futuristic environment",
+                            "golden hour warmth", "neon-lit urban setting", "minimalist aesthetic",
+                            "high-contrast look", "soft-focused dreamlike quality"
+                        ],
+                        label="Style/Atmosphere",
+                        value="cinematic atmosphere"
+                    )
+                build_custom_btn = gr.Button("🔨 Build Custom Prompt", variant="secondary")
+                custom_output = gr.Textbox(
+                    label="Your Custom Prompt",
+                    lines=3,
+                    interactive=True,
+                    show_copy_button=True
+                )
+
+        # Event handlers
+        analyze_btn.click(
+            fn=analyze_image_with_zephyr,
+            inputs=[image_input],
+            outputs=[analysis_output, scene_state]
+        )
+        sample_btn.click(
+            fn=generate_sample_prompts_with_zephyr,
+            inputs=[scene_state],
+            outputs=sample_prompts
+        )
+        optimize_btn.click(
+            fn=optimize_user_prompt_with_zephyr,
+            inputs=[user_idea, scene_state],
+            outputs=[optimized_prompt, model_status]
+        )
+        retry_btn.click(
+            fn=fallback_generate_prompt,
+            inputs=[user_idea, scene_state],
+            outputs=[optimized_prompt, model_status]
+        )
+        test_btn.click(
+            fn=test_basic_generation,
+            inputs=[],
+            outputs=[test_output]
+        )
+        debug_btn.click(
+            fn=get_debug_info,
+            inputs=[],
+            outputs=[debug_info]
+        )
+        refine_btn.click(
+            fn=refine_prompt_with_zephyr,
+            inputs=[optimized_prompt, feedback_input, chat_history_state, scene_state],
+            outputs=[optimized_prompt, chat_history_state]
+        )
+        # Update chat display when history changes
+        chat_history_state.change(
+            fn=lambda history: history,
+            inputs=[chat_history_state],
+            outputs=[chat_display]
+        )
+        build_custom_btn.click(
+            fn=build_custom_prompt,
+            inputs=[custom_foundation, subject_motion, scene_motion, camera_motion, style_motion],
+            outputs=[custom_output]
+        )
+    return demo
+
+# Launch the app
+if __name__ == "__main__":
+    print("🎬 Starting AI Video Prompt Generator with SARA LORA Adapter...")
+    print(f"📊 Status: {'GPU' if use_gpu else 'CPU'} Mode Enabled")
+    print("🔧 Loading models (this may take a few minutes)...")
+    try:
+        demo = create_interface()
+        print("✅ Interface created successfully!")
+        print("🚀 Launching application...")
+        demo.launch(
+            share=True,
+            server_name="0.0.0.0",
+            server_port=7860,
+            debug=True,
+            show_error=True
+        )
+    except Exception as e:
+        print(f"❌ Error launching app: {e}")
+        print("🔧 Make sure you have sufficient CPU resources and all dependencies installed.")
+        print("📦 Required packages:")
+        print("   pip install torch transformers gradio pillow accelerate bitsandbytes peft>=0.6.0")
+        # Alternative launch attempt
+        print("\n🔄 Attempting alternative launch...")
+        try:
+            # Try to install necessary dependencies
+            import subprocess
+            print("🔄 Installing/updating necessary dependencies...")
+            subprocess.call(["pip", "install", "-U", "transformers", "accelerate", "peft>=0.6.0", "huggingface_hub", "bitsandbytes"])
+            
+            demo = create_interface()
+            demo.launch(
+                share=False,
+                server_name="127.0.0.1",
+                server_port=7860,
+                debug=False
+            )
+        except Exception as e2:
+            print(f"❌ Alternative launch failed: {e2}")
+            print("\n💡 Troubleshooting tips:")
+            print("1. Ensure CPU resources are sufficient.")
+            print("2. Check CPU usage: top or htop")
+            print("3. Try reducing model precision: set torch_dtype=torch.float16")
+            print("4. Monitor memory usage: free -h")