Update app.py

app.py

@@ -50,7 +50,6 @@ def get_llama_pipeline(model_id: str, token: str):
     """
     if model_id in LLAMA_PIPELINES:
         return LLAMA_PIPELINES[model_id]
-
     tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=token)
     model = AutoModelForCausalLM.from_pretrained(
         model_id,
@@ -63,7 +62,6 @@ def get_llama_pipeline(model_id: str, token: str):
     LLAMA_PIPELINES[model_id] = text_pipeline
     return text_pipeline
 
-
 def get_musicgen_model(model_key: str = "facebook/musicgen-large"):
     """
     Returns a cached MusicGen model if available; otherwise, loads it.
@@ -71,7 +69,6 @@ def get_musicgen_model(model_key: str = "facebook/musicgen-large"):
     """
     if model_key in MUSICGEN_MODELS:
         return MUSICGEN_MODELS[model_key]
-
     model = MusicgenForConditionalGeneration.from_pretrained(model_key)
     processor = AutoProcessor.from_pretrained(model_key)
     device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -79,19 +76,16 @@ def get_musicgen_model(model_key: str = "facebook/musicgen-large"):
     MUSICGEN_MODELS[model_key] = (model, processor)
     return model, processor
 
-
 def get_tts_model(model_name: str = "tts_models/en/ljspeech/tacotron2-DDC"):
     """
     Returns a cached TTS model if available; otherwise, loads it.
     """
     if model_name in TTS_MODELS:
         return TTS_MODELS[model_name]
-
     tts_model = TTS(model_name)
     TTS_MODELS[model_name] = tts_model
     return tts_model
 
-
 # ---------------------------------------------------------------------
 # Script Generation Function
 # ---------------------------------------------------------------------
@@ -125,7 +119,7 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
         voice_script = "No voice-over script found."
         sound_design = "No sound design suggestions found."
         music_suggestions = "No music suggestions found."
-        # Voice-Over Script
+        # Extract sections if present
         if "Voice-Over Script:" in generated_text:
             parts = generated_text.split("Voice-Over Script:")
             voice_script_part = parts[1]
@@ -133,7 +127,6 @@ 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:")
             sound_design_part = parts[1]
@@ -141,7 +134,6 @@ 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()
@@ -149,7 +141,6 @@ def generate_script(user_prompt: str, model_id: str, token: str, duration: int):
     except Exception as e:
         return f"Error generating script: {e}", "", ""
 
-
 # ---------------------------------------------------------------------
 # Ad Promo Idea Generation Function
 # ---------------------------------------------------------------------
@@ -181,7 +172,6 @@ def generate_ad_promo_idea(user_prompt: str, model_id: str, token: str):
     except Exception as e:
         return f"Error generating ad promo idea: {e}"
 
-
 # ---------------------------------------------------------------------
 # Voice-Over Generation Function
 # ---------------------------------------------------------------------
@@ -202,7 +192,6 @@ def generate_voice(script: str, tts_model_name: str = "tts_models/en/ljspeech/ta
     except Exception as e:
         return f"Error generating voice: {e}"
 
-
 # ---------------------------------------------------------------------
 # Music Generation Function
 # ---------------------------------------------------------------------
@@ -229,7 +218,6 @@ def generate_music(prompt: str, audio_length: int):
     except Exception as e:
         return f"Error generating music: {e}"
 
-
 # ---------------------------------------------------------------------
 # Audio Blending with Duration Sync & Ducking
 # ---------------------------------------------------------------------
@@ -264,7 +252,6 @@ def blend_audio(voice_path: str, music_path: str, ducking: bool, duck_level: int
     except Exception as e:
         return f"Error blending audio: {e}"
 
-
 # ---------------------------------------------------------------------
 # Gradio Interface with Enhanced UI
 # ---------------------------------------------------------------------
@@ -284,19 +271,23 @@ with gr.Blocks(css="""
     }
     .header h1 {
         margin: 0;
-        font-size: 2.5rem;
+        font-size: 2.8rem;
     }
     .header p {
         font-size: 1.2rem;
     }
+    .instructions {
+        background-color: #2e2e2e;
+        border-radius: 8px;
+        padding: 1rem;
+        margin-bottom: 1rem;
+        font-size: 0.95rem;
+    }
     .gradio-container {
         background: #2e2e2e;
         border-radius: 10px;
         padding: 1rem;
-    }
-    .tab-title {
-        font-size: 1.1rem;
-        font-weight: bold;
+        margin-bottom: 1rem;
     }
     .footer {
         text-align: center;
@@ -305,6 +296,11 @@ with gr.Blocks(css="""
         padding: 1rem;
         color: #cccccc;
     }
+    .btn-clear {
+        margin-left: 1rem;
+        background: #ff5555;
+        color: #fff;
+    }
 """) as demo:
 
     # Custom Header
@@ -315,22 +311,23 @@ 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:
-    
-    - **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**: Generate custom music tracks with MusicGen.
-    - **Audio Blending**: Seamlessly combine voice and music with ducking options.
+    Welcome to **AI Ads Promo (Demo MVP)**! This platform leverages state-of-the-art AI models to help you generate creative advertising content.
+    Use the tabs below to generate:
+    - **Ad Promo Ideas**
+    - **Voice-Over Scripts**
+    - **Natural-Sounding Voice-Overs**
+    - **Custom Music Tracks**
+    - **Blended Audio Ads**
     """)
 
     with gr.Tabs():
-        # New Tab: Generate Ad Promo Idea
+        # Tab 1: Ad Promo Idea Generation
         with gr.Tab("💡 Ad Promo Idea"):
+            gr.Markdown("Enter a concept for your ad and let the system generate a creative ad promo idea with taglines and media suggestions.")
             with gr.Row():
                 ad_concept = gr.Textbox(
                     label="Ad Concept",
-                    placeholder="Enter your ad concept or idea...",
+                    placeholder="E.g., A vibrant summer sale for a trendy clothing brand...",
                     lines=2
                 )
             with gr.Row():
@@ -339,20 +336,24 @@ with gr.Blocks(css="""
                     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")
+            with gr.Row():
+                generate_ad_idea_button = gr.Button("Generate Ad Promo Idea", variant="primary")
+                clear_ad_idea = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
             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
             )
+            clear_ad_idea.click(fn=lambda: "", inputs=None, outputs=ad_idea_output)
 
-        # Step 1: Generate Script
+        # Tab 2: Script Generation
         with gr.Tab("📝 Script Generation"):
+            gr.Markdown("Generate a voice-over script along with sound design and music suggestions based on your promo idea.")
             with gr.Row():
                 user_prompt = gr.Textbox(
                     label="Promo Idea", 
-                    placeholder="E.g., A 30-second promo for a morning show...",
+                    placeholder="E.g., A 30-second energetic promo for a new product launch...",
                     lines=2
                 )
             with gr.Row():
@@ -368,19 +369,22 @@ with gr.Blocks(css="""
                     step=15, 
                     value=30
                 )
-            generate_script_button = gr.Button("Generate Script", variant="primary")
-            script_output = gr.Textbox(label="Generated Voice-Over Script", lines=5, interactive=False)
+            with gr.Row():
+                generate_script_button = gr.Button("Generate Script", variant="primary")
+                clear_script = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
+            script_output = gr.Textbox(label="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],
-                outputs=[script_output, sound_design_output, music_suggestion_output],
+                outputs=[script_output, sound_design_output, music_suggestion_output]
             )
+            clear_script.click(fn=lambda: ["", "", ""], inputs=None, outputs=[script_output, sound_design_output, music_suggestion_output])
 
-        # Step 2: Generate Voice
+        # Tab 3: Voice Synthesis
         with gr.Tab("🎤 Voice Synthesis"):
-            gr.Markdown("Generate a natural-sounding voice-over using Coqui TTS.")
+            gr.Markdown("Convert your generated script into a natural-sounding voice-over using Coqui TTS.")
             selected_tts_model = gr.Dropdown(
                 label="TTS Model",
                 choices=[
@@ -391,17 +395,19 @@ with gr.Blocks(css="""
                 value="tts_models/en/ljspeech/tacotron2-DDC",
                 multiselect=False
             )
-            generate_voice_button = gr.Button("Generate Voice-Over", variant="primary")
+            with gr.Row():
+                generate_voice_button = gr.Button("Generate Voice-Over", variant="primary")
+                clear_voice = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
             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],
-                outputs=voice_audio_output,
+                inputs=script_output, outputs=voice_audio_output
             )
+            clear_voice.click(fn=lambda: "", inputs=None, outputs=voice_audio_output)
 
-        # Step 3: Generate Music
+        # Tab 4: Music Production
         with gr.Tab("🎶 Music Production"):
-            gr.Markdown("Generate a custom music track using the **MusicGen Large** model.")
+            gr.Markdown("Generate a custom music track based on the suggestions using the MusicGen model.")
             audio_length = gr.Slider(
                 label="Music Length (tokens)",
                 minimum=128, 
@@ -410,17 +416,20 @@ with gr.Blocks(css="""
                 value=512,
                 info="Increase tokens for longer audio (inference time may vary)."
             )
-            generate_music_button = gr.Button("Generate Music", variant="primary")
+            with gr.Row():
+                generate_music_button = gr.Button("Generate Music", variant="primary")
+                clear_music = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
             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],
-                outputs=[music_output],
+                outputs=[music_output]
             )
+            clear_music.click(fn=lambda: "", inputs=None, outputs=music_output)
 
-        # Step 4: Blend Audio
+        # Tab 5: Audio Blending
         with gr.Tab("🎚️ Audio Blending"):
-            gr.Markdown("Blend your voice-over and music track. Music will be looped/truncated to match the voice duration. Enable ducking to lower the music during voice segments.")
+            gr.Markdown("Blend your voice-over and music track. Music will be adjusted to match the voice duration with an option to enable ducking.")
             ducking_checkbox = gr.Checkbox(label="Enable Ducking?", value=True)
             duck_level_slider = gr.Slider(
                 label="Ducking Level (dB attenuation)", 
@@ -429,13 +438,16 @@ with gr.Blocks(css="""
                 step=1, 
                 value=10
             )
-            blend_button = gr.Button("Blend Voice + Music", variant="primary")
+            with gr.Row():
+                blend_button = gr.Button("Blend Voice + Music", variant="primary")
+                clear_blend = gr.Button("Clear", variant="stop", elem_classes="btn-clear")
             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],
                 outputs=blended_output
             )
+            clear_blend.click(fn=lambda: "", inputs=None, outputs=blended_output)
 
     # Footer
     gr.Markdown("""
@@ -447,7 +459,6 @@ with gr.Blocks(css="""
     </div>
     """)
     
-    # Visitor Badge
     gr.HTML("""
     <div style="text-align: center; margin-top: 1rem;">
         <a href="https://visitorbadge.io/status?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2FBils%2Fradiogold">