Update app.py

app.py

@@ -39,7 +39,6 @@ def clean_text(text: str) -> str:
     """
     Removes undesired characters (e.g., asterisks) that might not be recognized by the model's vocabulary.
     """
-    # Remove all asterisks.
     return re.sub(r'\*', '', text)
 
 # ---------------------------------------------------------------------
@@ -75,7 +74,6 @@ def get_musicgen_model(model_key: str = "facebook/musicgen-large"):
 
     model = MusicgenForConditionalGeneration.from_pretrained(model_key)
     processor = AutoProcessor.from_pretrained(model_key)
-
     device = "cuda" if torch.cuda.is_available() else "cpu"
     model.to(device)
     MUSICGEN_MODELS[model_key] = (model, processor)
@@ -105,7 +103,6 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
     """
     try:
         text_pipeline = get_llama_pipeline(model_id, token)
-
         system_prompt = (
             "You are an expert radio imaging producer specializing in sound design and music. "
             f"Based on the user's concept and the selected duration of {duration} seconds, produce the following: "
@@ -114,7 +111,6 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
             "3. Music styles or track recommendations. Prefix this section with 'Music Suggestions:'."
         )
         combined_prompt = f"{system_prompt}\nUser concept: {user_prompt}\nOutput:"
-
         with torch.inference_mode():
             result = text_pipeline(
                 combined_prompt,
@@ -122,16 +118,13 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
                 do_sample=True,
                 temperature=0.8
             )
-
         generated_text = result[0]["generated_text"]
         if "Output:" in generated_text:
             generated_text = generated_text.split("Output:")[-1].strip()
-
         # Default placeholders
         voice_script = "No voice-over script found."
         sound_design = "No sound design suggestions found."
         music_suggestions = "No music suggestions found."
-
         # Voice-Over Script
         if "Voice-Over Script:" in generated_text:
             parts = generated_text.split("Voice-Over Script:")
@@ -140,7 +133,6 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
                 voice_script = voice_script_part.split("Sound Design Suggestions:")[0].strip()
             else:
                 voice_script = voice_script_part.strip()
-
         # Sound Design
         if "Sound Design Suggestions:" in generated_text:
             parts = generated_text.split("Sound Design Suggestions:")
@@ -149,18 +141,47 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
                 sound_design = sound_design_part.split("Music Suggestions:")[0].strip()
             else:
                 sound_design = sound_design_part.strip()
-
         # Music Suggestions
         if "Music Suggestions:" in generated_text:
             parts = generated_text.split("Music Suggestions:")
             music_suggestions = parts[1].strip()
-
         return voice_script, sound_design, music_suggestions
-
     except Exception as e:
         return f"Error generating script: {e}", "", ""
 
 
+# ---------------------------------------------------------------------
+# Ad Promo Idea Generation Function
+# ---------------------------------------------------------------------
+@spaces.GPU(duration=100)
+def generate_ad_promo_idea(user_prompt: str, model_id: str, token: str):
+    """
+    Generates a creative ad promo idea based on the user's concept.
+    Returns a string containing the ad promo idea.
+    """
+    try:
+        text_pipeline = get_llama_pipeline(model_id, token)
+        system_prompt = (
+            "You are a creative advertising strategist. "
+            "Generate a unique and engaging ad promo idea based on the following concept. "
+            "Include creative angles, potential taglines, and media suggestions."
+        )
+        combined_prompt = f"{system_prompt}\nConcept: {user_prompt}\nAd Promo Idea:"
+        with torch.inference_mode():
+            result = text_pipeline(
+                combined_prompt,
+                max_new_tokens=150,
+                do_sample=True,
+                temperature=0.8
+            )
+        generated_text = result[0]["generated_text"]
+        if "Ad Promo Idea:" in generated_text:
+            generated_text = generated_text.split("Ad Promo Idea:")[-1].strip()
+        return generated_text
+    except Exception as e:
+        return f"Error generating ad promo idea: {e}"
+
+
 # ---------------------------------------------------------------------
 # Voice-Over Generation Function
 # ---------------------------------------------------------------------
@@ -173,17 +194,11 @@ def generate_voice(script: str, tts_model_name: str = "tts_models/en/ljspeech/ta
     try:
         if not script.strip():
             return "Error: No script provided."
-
-        # Clean the script to remove special characters (e.g., asterisks) that may produce warnings
         cleaned_script = clean_text(script)
-
         tts_model = get_tts_model(tts_model_name)
-
-        # Generate and save voice
         output_path = os.path.join(tempfile.gettempdir(), "voice_over.wav")
         tts_model.tts_to_file(text=cleaned_script, file_path=output_path)
         return output_path
-
     except Exception as e:
         return f"Error generating voice: {e}"
 
@@ -200,24 +215,17 @@ def generate_music(prompt: str, audio_length: int):
     try:
         if not prompt.strip():
             return "Error: No music suggestion provided."
-
         model_key = "facebook/musicgen-large"
         musicgen_model, musicgen_processor = get_musicgen_model(model_key)
-
         device = "cuda" if torch.cuda.is_available() else "cpu"
         inputs = musicgen_processor(text=[prompt], padding=True, return_tensors="pt").to(device)
-
         with torch.inference_mode():
             outputs = musicgen_model.generate(**inputs, max_new_tokens=audio_length)
-
         audio_data = outputs[0, 0].cpu().numpy()
         normalized_audio = (audio_data / max(abs(audio_data)) * 32767).astype("int16")
-
         output_path = os.path.join(tempfile.gettempdir(), "musicgen_large_generated_music.wav")
         write(output_path, 44100, normalized_audio)
-
         return output_path
-
     except Exception as e:
         return f"Error generating music: {e}"
 
@@ -229,42 +237,30 @@ def generate_music(prompt: str, audio_length: int):
 def blend_audio(voice_path: str, music_path: str, ducking: bool, duck_level: int = 10):
     """
     Blends two audio files (voice and music).
-    1. If music < voice, loops the music until it meets/exceeds the voice duration.
-    2. If music > voice, trims music to the voice duration.
-    3. If ducking=True, the music is attenuated by 'duck_level' dB while the voice is playing.
     Returns the file path to the blended .wav file.
     """
     try:
         if not os.path.isfile(voice_path) or not os.path.isfile(music_path):
             return "Error: Missing audio files for blending."
-
         voice = AudioSegment.from_wav(voice_path)
         music = AudioSegment.from_wav(music_path)
-
-        voice_len = len(voice)  # in milliseconds
-        music_len = len(music)  # in milliseconds
-
-        # Loop music if it's shorter than the voice
+        voice_len = len(voice)
+        music_len = len(music)
         if music_len < voice_len:
             looped_music = AudioSegment.empty()
             while len(looped_music) < voice_len:
                 looped_music += music
             music = looped_music
-
-        # Trim music if it's longer than the voice
         if len(music) > voice_len:
             music = music[:voice_len]
-
         if ducking:
             ducked_music = music - duck_level
             final_audio = ducked_music.overlay(voice)
         else:
             final_audio = music.overlay(voice)
-
         output_path = os.path.join(tempfile.gettempdir(), "blended_output.wav")
         final_audio.export(output_path, format="wav")
         return output_path
-
     except Exception as e:
         return f"Error blending audio: {e}"
 
@@ -321,13 +317,36 @@ with gr.Blocks(css="""
     gr.Markdown("""
     Welcome to **AI Ads Promo (Demo MVP)**! This platform leverages state-of-the-art AI models to help you generate:
     
-    - **Script**: Generate a compelling voice-over script with LLaMA.
+    - **Ad Promo Ideas**: Generate creative ad concepts.
+    - **Script**: Produce a compelling voice-over script with LLaMA.
     - **Voice Synthesis**: Create natural-sounding voice-overs using Coqui TTS.
-    - **Music Production**: Produce custom music tracks with MusicGen.
-    - **Audio Blending**: Seamlessly blend voice and music with options for ducking.
+    - **Music Production**: Generate custom music tracks with MusicGen.
+    - **Audio Blending**: Seamlessly combine voice and music with ducking options.
     """)
 
     with gr.Tabs():
+        # New Tab: Generate Ad Promo Idea
+        with gr.Tab("💡 Ad Promo Idea"):
+            with gr.Row():
+                ad_concept = gr.Textbox(
+                    label="Ad Concept",
+                    placeholder="Enter your ad concept or idea...",
+                    lines=2
+                )
+            with gr.Row():
+                llama_model_id_idea = gr.Textbox(
+                    label="LLaMA Model ID", 
+                    value="meta-llama/Meta-Llama-3-8B-Instruct", 
+                    placeholder="Enter a valid Hugging Face model ID"
+                )
+            generate_ad_idea_button = gr.Button("Generate Ad Promo Idea", variant="primary")
+            ad_idea_output = gr.Textbox(label="Generated Ad Promo Idea", lines=5, interactive=False)
+            generate_ad_idea_button.click(
+                fn=lambda concept, model_id: generate_ad_promo_idea(concept, model_id, HF_TOKEN),
+                inputs=[ad_concept, llama_model_id_idea],
+                outputs=ad_idea_output
+            )
+
         # Step 1: Generate Script
         with gr.Tab("📝 Script Generation"):
             with gr.Row():
@@ -353,7 +372,6 @@ with gr.Blocks(css="""
             script_output = gr.Textbox(label="Generated Voice-Over Script", lines=5, interactive=False)
             sound_design_output = gr.Textbox(label="Sound Design Suggestions", lines=3, interactive=False)
             music_suggestion_output = gr.Textbox(label="Music Suggestions", lines=3, interactive=False)
-
             generate_script_button.click(
                 fn=lambda user_prompt, model_id, dur: generate_script(user_prompt, model_id, HF_TOKEN, dur),
                 inputs=[user_prompt, llama_model_id, duration],
@@ -375,7 +393,6 @@ with gr.Blocks(css="""
             )
             generate_voice_button = gr.Button("Generate Voice-Over", variant="primary")
             voice_audio_output = gr.Audio(label="Voice-Over (WAV)", type="filepath")
-
             generate_voice_button.click(
                 fn=lambda script, tts_model: generate_voice(script, tts_model),
                 inputs=[script_output, selected_tts_model],
@@ -395,7 +412,6 @@ with gr.Blocks(css="""
             )
             generate_music_button = gr.Button("Generate Music", variant="primary")
             music_output = gr.Audio(label="Generated Music (WAV)", type="filepath")
-
             generate_music_button.click(
                 fn=lambda music_suggestion, length: generate_music(music_suggestion, length),
                 inputs=[music_suggestion_output, audio_length],
@@ -415,7 +431,6 @@ with gr.Blocks(css="""
             )
             blend_button = gr.Button("Blend Voice + Music", variant="primary")
             blended_output = gr.Audio(label="Final Blended Output (WAV)", type="filepath")
-
             blend_button.click(
                 fn=blend_audio,
                 inputs=[voice_audio_output, music_output, ducking_checkbox, duck_level_slider],