Update app.py

app.py

@@ -1,272 +1,1508 @@
-import os
-import tempfile
-import torch
 import gradio as gr
-from huggingface_hub import hf_hub_download, snapshot_download
-import spaces
-
-# Download models from HuggingFace Hub on startup
-def download_models():
-    """Download all required model files from HuggingFace Hub."""
-    cache_dir = os.environ.get("HF_HOME", os.path.expanduser("/tmp"))
-    model_dir = os.path.join(cache_dir, "heartmula_models")
-
-    if not os.path.exists(model_dir):
-        os.makedirs(model_dir, exist_ok=True)
-
-    # Download HeartMuLaGen (tokenizer and gen_config)
-    print("Downloading HeartMuLaGen files...")
-    for filename in ["tokenizer.json", "gen_config.json"]:
-        hf_hub_download(
-            repo_id="HeartMuLa/HeartMuLaGen",
-            filename=filename,
-            local_dir=model_dir,
-        )
+import requests
+import json
+import os
+from groq import Groq
 
-    # Download HeartMuLa-oss-3B
-    print("Downloading HeartMuLa-oss-3B...")
-    snapshot_download(
-        repo_id="HeartMuLa/HeartMuLa-oss-3B",
-        local_dir=os.path.join(model_dir, "HeartMuLa-oss-3B"),
-    )
+# ============================================================
+# 🎵 SOMA Music Studio - HeartMuLa Optimized Edition
+# MiniMax Music 2.5 + HeartMuLa Style Guide + Comic Classic Theme
+# ============================================================
 
-    # Download HeartCodec-oss
-    print("Downloading HeartCodec-oss...")
-    snapshot_download(
-        repo_id="HeartMuLa/HeartCodec-oss",
-        local_dir=os.path.join(model_dir, "HeartCodec-oss"),
-    )
+# HeartMuLa 권장 구조 태그 (공식 문서 기반)
+STRUCTURE_TAGS = [
+    "[Intro]", "[Verse]", "[Prechorus]", "[Chorus]", "[Bridge]",
+    "[Interlude]", "[Hook]", "[Outro]", "[Inst]", "[Solo]"
+]
 
-    print("All models downloaded successfully!")
-    return model_dir
-
-from heartlib import HeartMuLaGenPipeline
-
-model_dir = download_models()
-
-# Determine device and dtype
-if torch.cuda.is_available():
-    device = torch.device("cuda")
-    dtype = torch.bfloat16
-else:
-    device = torch.device("cpu")
-    dtype = torch.float32
-
-print(f"Loading pipeline on {device} with {dtype}...")
-pipe = HeartMuLaGenPipeline.from_pretrained(
-    model_dir,
-    device=device,
-    dtype=dtype,
-    version="3B",
-)
-print("Pipeline loaded successfully!")
-
-
-@spaces.GPU(duration=130)
-def generate_music(
-    lyrics: str,
-    tags: str,
-    max_duration_seconds: int,
-    temperature: float,
-    topk: int,
-    cfg_scale: float,
-    progress=gr.Progress(track_tqdm=True),
-):
-    """Generate music from lyrics and tags."""
-    if not lyrics.strip():
-        raise gr.Error("Please enter some lyrics!")
-
-    if not tags.strip():
-        raise gr.Error("Please enter at least one tag!")
-
-    # Create a temporary file for output
-    with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as f:
-        output_path = f.name
-
-    max_audio_length_ms = max_duration_seconds * 1000
-
-    with torch.no_grad():
-        pipe(
-            {
-                "lyrics": lyrics,
-                "tags": tags,
-            },
-            max_audio_length_ms=max_audio_length_ms,
-            save_path=output_path,
-            topk=topk,
-            temperature=temperature,
-            cfg_scale=cfg_scale,
-        )
+# HeartMuLa 스타일 태그 가이드 (콤마 구분, 공백 없음)
+HEARTMULA_TAG_GUIDE = """
+## 🎼 HeartMuLa Tag Format Guide
+
+### STRUCTURE TAGS (가사 내 사용):
+[Intro], [Verse], [Prechorus], [Chorus], [Bridge], [Interlude], [Hook], [Outro], [Inst], [Solo]
+
+### STYLE TAGS FORMAT (콤마 구분, 공백 없음):
+```
+piano,happy,wedding,synthesizer,romantic
+guitar,sad,ballad,strings,emotional
+drums,energetic,rock,electric_guitar,powerful
+```
+
+### RECOMMENDED TAG CATEGORIES:
+**Instruments:** piano,guitar,drums,bass,synthesizer,strings,violin,cello,trumpet,saxophone,flute,harp,organ
+**Mood:** happy,sad,romantic,energetic,calm,melancholic,uplifting,dark,dreamy,nostalgic,powerful,peaceful
+**Genre:** pop,rock,jazz,classical,electronic,folk,blues,r&b,hip_hop,disco,ballad,cinematic
+**Tempo:** fast,slow,moderate,upbeat,relaxed
+**Occasion:** wedding,party,meditation,workout,study,sleep,travel
+
+### EXAMPLE COMBINATIONS:
+- K-Pop Dance: `synthesizer,drums,energetic,pop,upbeat,powerful`
+- Jazz Ballad: `piano,saxophone,romantic,jazz,slow,dreamy`
+- Epic Cinematic: `strings,orchestra,powerful,cinematic,dramatic,epic`
+- Lo-Fi Chill: `piano,guitar,calm,lo_fi,relaxed,dreamy`
+"""
+
+# MiniMax Music 2.0 핵심 기능 가이드 (강화)
+MINIMAX_MUSIC_GUIDE = """
+## MiniMax Music 2.5 Core Capabilities:
+
+### 1. DYNAMIC VOCALS - Mastery Over Diverse Singing Styles
+- Human-like vocal timbre with professional singing techniques
+- Precise control over phrasing, rhythm, and breath
+- One voice can switch between multiple styles
+- Supports: Pop, Jazz, Blues, Rock, Folk, Electronic, Urban, Disco
+- Special modes: Male-female duets, A Cappella (pure vocals)
 
-    return output_path
+### 2. CATCHY MELODIES & INSTRUMENT CONTROL  
+- Structurally complete songs: Verse → Chorus → Bridge (up to 5 minutes)
+- Memorable, instantly captivating melodies
+- Independent control of individual instruments
+- Instruments: Piano, Guitar, Bass, Drums, Saxophone, Trumpet, Synths, Strings
 
+### 3. PROFESSIONAL-GRADE AUDIO
+- Enhanced vocal track texture
+- Spatial presence of instruments
+- Immersive listening experience
+- Film-grade monologue soundtracks
 
-# Example lyrics
-EXAMPLE_LYRICS = """[Intro]
+### 4. PROMPT WRITING BEST PRACTICES (HeartMuLa Optimized)
+- Use specific instrument names
+- Describe vocal emotions precisely
+- Specify singing techniques (breathy, powerful, smooth, raspy)
+- Use tempo (BPM), key when needed
+- Tags should be comma-separated WITHOUT spaces
+"""
+
+# HeartMuLa 권장 가사 구조 예시
+HEARTMULA_LYRICS_STRUCTURE = """
+## 📝 HeartMuLa Recommended Lyrics Structure
+
+### OPTIMAL FORMAT:
+```
+[Intro]
 
 [Verse]
-The sun creeps in across the floor
-I hear the traffic outside the door
-The coffee pot begins to hiss
-It is another morning just like this
+First verse lyrics here
+Second line of first verse
+Third line continues story
+Fourth line builds emotion
 
 [Prechorus]
-The world keeps spinning round and round
-Feet are planted on the ground
-I find my rhythm in the sound
+Building tension here
+Leading to the chorus
 
 [Chorus]
-Every day the light returns
-Every day the fire burns
-We keep on walking down this street
-Moving to the same steady beat
-It is the ordinary magic that we meet
+Main hook and memorable melody
+Most important part of song
+Repeat this section 2-3 times
+Make it singable and catchy
 
 [Verse]
-The hours tick deeply into noon
-Chasing shadows, chasing the moon
-Work is done and the lights go low
-Watching the city start to glow
+Second verse develops story
+New information revealed
+Emotional progression continues
+Building toward bridge
 
 [Bridge]
-It is not always easy, not always bright
-Sometimes we wrestle with the night
-But we make it to the morning light
+Contrast section here
+Different melody or perspective
+Emotional peak moment
 
 [Chorus]
-Every day the light returns
-Every day the fire burns
-We keep on walking down this street
-Moving to the same steady beat
+Main hook repeated
+With possible variations
+Final emotional release
 
 [Outro]
-Just another day
-Every single day"""
-
-EXAMPLE_TAGS = "piano,happy,uplifting,pop"
-
-# Build the Gradio interface
-with gr.Blocks(
-    title="HeartMuLa Music Generator",
-) as demo:
-    gr.Markdown(
-        """
-        # HeartMuLa Music Generator
-
-        Generate music from lyrics and tags using [HeartMuLa](https://github.com/HeartMuLa/heartlib),
-        an open-source music foundation model.
-
-        **Instructions:**
-        1. Enter your lyrics with structure tags like `[Verse]`, `[Chorus]`, `[Bridge]`, etc.
-        2. Add comma-separated tags describing the music style (e.g., `piano,happy,romantic`)
-        3. Adjust generation parameters as needed
-        4. Click "Generate Music" and wait for your song!
-
-        *Note: Generation can take several minutes depending on the duration.*
-        """
-    )
+Closing the song
+Fading or resolving
+```
 
-    with gr.Row():
-        with gr.Column(scale=1):
-            lyrics_input = gr.Textbox(
-                label="Lyrics",
-                placeholder="Enter lyrics with structure tags like [Verse], [Chorus], etc.",
-                lines=20,
-                value=EXAMPLE_LYRICS,
-            )
+### KEY RULES:
+1. [Chorus] appears at least 2-3 times
+2. [Prechorus] builds tension before [Chorus]
+3. [Bridge] provides contrast before final [Chorus]
+4. Each [Verse] should progress the story
+5. [Intro] and [Outro] frame the song
+"""
 
-            tags_input = gr.Textbox(
-                label="Tags",
-                placeholder="piano,happy,romantic,synthesizer",
-                value=EXAMPLE_TAGS,
-                info="Comma-separated tags describing the music style",
-            )
+# 예제 프롬프트 (HeartMuLa + MiniMax 최적화)
+EXAMPLE_PROMPTS = {
+    "🎤 A Cappella (아카펠라)": {
+        "prompt": "A cappella arrangement with pure vocal harmonies, no instrumental accompaniment. Features a lead soprano voice with rich layered backing vocals creating lush harmonies. Gentle humming bass line, rhythmic vocal percussion, and ethereal 'ooh' and 'aah' pads. The vocals blend seamlessly with precise tuning and warm reverb, creating a refreshing, meditative atmosphere. 70 BPM, peaceful and soothing mood.",
+        "tags": "acappella,vocals,harmony,peaceful,soothing,choir",
+        "description": "순수 보컬만으로 풍부한 멜로디"
+    },
+    "👥 Group Harmony (그룹 하모니)": {
+        "prompt": "Powerful group vocal anthem featuring a lead female voice with layered choir harmonies. Rich unison sections build to explosive harmonic splits in the chorus. Features call-and-response patterns, anthemic 'oh-oh-oh' chants, and soaring group harmonies. Modern pop production with punchy drums, driving bass, synth hooks, and brass stabs. 118 BPM, empowering and triumphant energy.",
+        "tags": "choir,harmony,pop,drums,bass,synthesizer,brass,powerful,anthem",
+        "description": "파워풀한 그룹 보컬 앤썸"
+    },
+    "🎷 Jazz Duet (재즈 듀엣)": {
+        "prompt": "Intimate jazz duet featuring conversational interplay between a warm male baritone and a silky female alto voice. Dynamic intensity variations with seamless transitions between lead vocals. Accompanied by brushed jazz drums, walking upright bass, and gentle piano comping. Saxophone solo in the bridge. 95 BPM, late-night jazz club atmosphere with warm analog sound.",
+        "tags": "jazz,piano,bass,drums,saxophone,romantic,intimate,duet",
+        "description": "남녀 보컬의 재즈 듀엣"
+    },
+    "🎸 Multi-Style (멀티스타일)": {
+        "prompt": "Showcase track demonstrating one female voice transitioning through three distinct styles: Starting with energetic Jump Blues featuring powerful belting and brass stabs, transitioning to aggressive Rock with distorted guitars and raspy vocals, finally morphing into sleek Electronic with auto-tuned vocals and pulsing synths. 120 BPM with tempo shifts between sections.",
+        "tags": "blues,rock,electronic,guitar,synthesizer,brass,dynamic,powerful",
+        "description": "스타일 전환 쇼케이스"
+    },
+    "🌃 Urban Chill (어반 칠)": {
+        "prompt": "Contemporary urban R&B track with a cool, laid-back vibe. Features a smooth male vocal with subtle Auto-Tune enhancement and breathy delivery. Trap-influenced 808 bass, crisp hi-hats with intricate patterns, ambient synth pads, and soft piano chords. Spacious production with heavy reverb and delay. 85 BPM, modern and sophisticated sound.",
+        "tags": "r&b,808,synthesizer,piano,chill,urban,modern,smooth",
+        "description": "어반 R&B 무드"
+    },
+    "🎹 Jazz Club (재즈 클럽)": {
+        "prompt": "Live jazz ensemble performance capturing the essence of Blue Note club. Instruments enter in perfect sequence: brushed drums set the groove, walking bass joins in, piano adds sophisticated chord voicings, then saxophone takes the melody. Trumpet and trombone provide punchy brass accents. Extended saxophone solo with bebop-style improvisation. 140 BPM swing feel.",
+        "tags": "jazz,piano,bass,drums,saxophone,trumpet,trombone,live,swing",
+        "description": "라이브 재즈 클럽"
+    },
+    "🪩 Retro Disco (레트로 디스코)": {
+        "prompt": "Vibrant disco track channeling the golden age of 80s dance music. Features a powerful female diva vocal with dynamic range and soulful ad-libs. Classic instrumentation: four-on-the-floor kick drum, funky slap bass, rhythmic guitar scratches, lush string arrangements, and bright brass stabs. Warm analog tape saturation. 120 BPM, euphoric and nostalgic energy.",
+        "tags": "disco,bass,guitar,strings,brass,drums,retro,funky,dance",
+        "description": "80년대 디스코"
+    },
+    "🎬 Film Score (영화 스코어)": {
+        "prompt": "Cinematic monologue soundtrack with layered emotional progression. A contemplative male voice delivers poetic narration over evolving orchestral arrangement. Begins with solo piano and soft strings, gradually building with French horn and cello. Atmospheric sound design with ocean waves and distant thunder. 60 BPM, deeply moving and introspective.",
+        "tags": "orchestral,piano,strings,cello,horn,cinematic,emotional,dramatic",
+        "description": "시네마틱 스코어"
+    },
+    "🎵 K-Pop Dance (K-Pop 댄스)": {
+        "prompt": "High-energy K-Pop dance track with a bright, clear female vocal and polished production. Catchy hook melody that's instantly memorable. Driving beat with punchy kicks, snappy snares, and intricate hi-hat programming. Layered synth hooks, powerful brass hits, and EDM-style buildups to explosive drops. 128 BPM, confident and empowering energy.",
+        "tags": "kpop,synthesizer,drums,bass,brass,electronic,energetic,dance,pop",
+        "description": "고에너지 K-Pop"
+    },
+    "🎻 Orchestral Ballad (오케스트라 발라드)": {
+        "prompt": "Sweeping orchestral ballad with an emotional female soprano voice. Begins intimately with solo piano, gradually introducing strings section - first violins, then violas and cellos. French horn provides warm counter-melodies. Builds to a full orchestral climax with timpani rolls and brass fanfares. 65 BPM, epic yet intimate.",
+        "tags": "orchestral,piano,violin,cello,horn,strings,ballad,emotional,epic",
+        "description": "오케스트라 발라드"
+    },
+    "🔥 HeartMuLa Default (기본 예시)": {
+        "prompt": "Uplifting pop song with piano and synthesizer leads, happy and romantic mood. Features a clear female vocal with emotional delivery, supported by gentle drums and warm bass. Catchy melody in the chorus with memorable hooks. Clean production with balanced mix. 110 BPM, wedding-appropriate joyful atmosphere.",
+        "tags": "piano,happy,wedding,synthesizer,romantic",
+        "description": "HeartMuLa 공식 예시"
+    }
+}
 
-            with gr.Accordion("Advanced Settings", open=False):
-                max_duration = gr.Slider(
-                    minimum=30,
-                    maximum=240,
-                    value=120,
-                    step=10,
-                    label="Max Duration (seconds)",
-                    info="Maximum length of generated audio",
-                )
+# SOMA 에이전트 - 가사 생성 (HeartMuLa 최적화)
+LYRICS_AGENTS = {
+    "lyricist": f"""You are a master lyricist optimized for HeartMuLa and MiniMax Music generation.
 
-                temperature = gr.Slider(
-                    minimum=0.1,
-                    maximum=2.0,
-                    value=1.0,
-                    step=0.1,
-                    label="Temperature",
-                    info="Higher = more creative, Lower = more consistent",
-                )
+{MINIMAX_MUSIC_GUIDE}
 
-                topk = gr.Slider(
-                    minimum=1,
-                    maximum=100,
-                    value=50,
-                    step=1,
-                    label="Top-K",
-                    info="Number of top tokens to sample from",
-                )
+{HEARTMULA_LYRICS_STRUCTURE}
 
-                cfg_scale = gr.Slider(
-                    minimum=1.0,
-                    maximum=3.0,
-                    value=1.5,
-                    step=0.1,
-                    label="CFG Scale",
-                    info="Classifier-free guidance scale",
-                )
+Your task: Create powerful, memorable lyrics with OPTIMAL HeartMuLa tag placement.
 
-            generate_btn = gr.Button("Generate Music", variant="primary", size="lg")
+CRITICAL RULES:
+1. Use HeartMuLa structure tags: [Intro], [Verse], [Prechorus], [Chorus], [Bridge], [Interlude], [Hook], [Outro], [Inst], [Solo]
+2. [Prechorus] (not [Pre Chorus]) - HeartMuLa format
+3. Ensure [Chorus] is the most memorable, singable part
+4. Include [Prechorus] to build tension before [Chorus]
+5. Add [Bridge] for emotional contrast
+6. Minimum 6-8 sections for quality generation
 
-        with gr.Column(scale=1):
-            audio_output = gr.Audio(
-                label="Generated Music",
-                type="filepath",
-            )
+Write lyrics that create the BEST POSSIBLE foundation for high-quality music generation.""",
+
+    "producer": f"""You are a music producer specializing in song structure optimization for AI music generation.
+
+{HEARTMULA_LYRICS_STRUCTURE}
+
+Your task: Analyze and OPTIMIZE the tag structure for MAXIMUM musical impact.
+
+CRITICAL OPTIMIZATION RULES:
+1. Use HeartMuLa tags: [Intro], [Verse], [Prechorus], [Chorus], [Bridge], [Interlude], [Hook], [Outro]
+2. Verify the structure follows genre conventions
+3. Ensure proper tag sequence (Verse→Prechorus→Chorus)
+4. Check that [Chorus] appears at least 2-3 times
+5. Verify [Bridge] provides contrast before final chorus
+6. Balance repetition and variation
+7. Ensure song length is appropriate (8-12 sections)
+
+Output the restructured lyrics with OPTIMAL HeartMuLa tag placement.""",
+
+    "emotion_director": f"""You are an emotion director for music production.
+
+{HEARTMULA_LYRICS_STRUCTURE}
+
+Your task: Enhance emotional impact through STRATEGIC tag content.
+
+EMOTIONAL MAPPING BY TAG:
+- [Intro]: Intrigue, anticipation
+- [Verse]: Storytelling, building connection  
+- [Prechorus]: Rising tension, excitement
+- [Chorus]: Peak emotion, catharsis
+- [Bridge]: Vulnerability, reflection, contrast
+- [Interlude]: Breathing space, transition
+- [Outro]: Resolution, lingering feeling
+
+OPTIMIZATION RULES:
+1. Each [Verse] should progress emotionally
+2. [Chorus] must deliver the strongest emotional punch
+3. [Bridge] should offer new emotional perspective
+4. [Prechorus] should create anticipation for [Chorus]
+5. Ensure dynamic contrast between sections
+
+Enhance the lyrics for MAXIMUM emotional resonance.""",
+
+    "final_editor": f"""You are the final editor for HeartMuLa/MiniMax Music production.
+
+{HEARTMULA_LYRICS_STRUCTURE}
+
+Your task: Output PERFECTLY FORMATTED, production-ready lyrics.
+
+CRITICAL OUTPUT RULES:
+1. Output ONLY the actual lyrics with structure tags
+2. Use HeartMuLa tags EXACTLY:
+   [Intro], [Verse], [Prechorus], [Chorus], [Bridge], [Interlude], [Hook], [Outro], [Inst], [Solo]
+3. DO NOT include English translations in parentheses
+4. DO NOT include explanations, descriptions, or markdown
+5. DO NOT include lines starting with * or >
+6. For [Inst] or [Solo] sections, write ONLY the tag
+7. Ensure MINIMUM 6-8 different sections
+8. Verify [Chorus] appears at least 2 times
+
+CORRECT FORMAT EXAMPLE:
+[Intro]
+
+[Verse]
+새벽 안개 속에 숨은 우리
+차가운 바람도 우리 발걸음에 무릎 꿇어
+
+[Prechorus]
+손을 잡고 빛을 꺼내
+
+[Chorus]
+우린 빛나는 별이 되어
+우리 목소리로 세상을 뒤집어
+Yeah yeah yeah
+
+[Verse]
+DNA에 새겨진 창조는 멈추지 않아
+
+[Prechorus]
+눈부신 파도를 타고
+
+[Chorus]
+우린 빛나는 별이 되어
+우리 목소리로 ���상을 뒤집어
+
+[Bridge]
+무지개가 흐르는 밤
+우리의 꿈은 불멸
+
+[Chorus]
+우린 빛나는 별이 되어
+새벽을 깨워
+
+[Outro]
+내가 만든 노래 영원히 타오른다
+
+OUTPUT ONLY CLEAN LYRICS WITH OPTIMAL TAG STRUCTURE."""
+}
+
+# SOMA 에이전트 - 프롬프트 증강 (HeartMuLa 최적화)
+PROMPT_AGENTS = {
+    "genre_specialist": f"""You are a music genre specialist for HeartMuLa and MiniMax Music.
+
+{MINIMAX_MUSIC_GUIDE}
+
+Analyze the input and identify:
+- Main genre and sub-genres
+- Era influences and regional styles
+- Specific production techniques for the genre
+- Instrument combinations that work best
+
+Output detailed genre characteristics optimized for music generation.""",
+
+    "sound_designer": """You are a sound designer for HeartMuLa and MiniMax Music.
+
+Define the complete sonic palette:
+- Specific instruments: Piano, Guitar (acoustic/electric), Bass (upright/electric/808), Drums (acoustic/electronic/brushed), Synths (pad/lead/bass), Brass (saxophone/trumpet/trombone), Strings (violin/viola/cello)
+- Drum patterns: four-on-the-floor, trap hi-hats, brushed jazz, etc.
+- Bass characteristics: walking bass, 808 sub-bass, slap bass, etc.
+- Atmospheric elements: reverb type, delay, spatial width
+- Production style: analog warmth, modern crisp, lo-fi, etc.
+
+Be extremely specific - both models can control individual instruments.""",
+
+    "vocal_director": """You are a vocal director for HeartMuLa and MiniMax Music.
+
+Both models excel at:
+- Human-like vocal timbre
+- Multiple singing styles from one voice
+- Male-female duets with conversational interplay
+- A Cappella with layered harmonies
+- Group/Choir with rich harmonic layers
+
+Define:
+- Voice type: male/female, age range, tone (warm/bright/husky/clear)
+- Singing technique: belting, falsetto, breathy, raspy, smooth
+- Vocal processing: reverb, delay, layers
+- Delivery style: confident, vulnerable, aggressive, intimate
+- For duets: describe each voice and their interaction
+- For A Cappella: describe harmony parts
+- For Group/Choir: describe layered vocals, unison sections""",
+
+    "tag_generator": f"""You are a style tag generator for HeartMuLa.
+
+{HEARTMULA_TAG_GUIDE}
+
+Your task: Generate OPTIMIZED comma-separated tags (NO SPACES between tags).
+
+RULES:
+1. Tags must be comma-separated WITHOUT spaces: tag1,tag2,tag3
+2. Use lowercase with underscores for multi-word: electric_guitar, hip_hop
+3. Include 5-8 tags covering: instrument, mood, genre, tempo
+4. Be specific: "piano" not "keyboard", "808" not "bass"
+5. Match the musical style described
+
+OUTPUT FORMAT (example):
+piano,synthesizer,happy,pop,upbeat,romantic,energetic
+
+Output ONLY the comma-separated tags, nothing else.""",
+
+    "prompt_synthesizer": f"""You are the final prompt synthesizer for HeartMuLa/MiniMax Music.
+
+{MINIMAX_MUSIC_GUIDE}
+
+Combine all inputs into ONE cohesive production prompt:
+- 150-200 words in English
+- Include: genre, specific BPM, instruments with details, vocal characteristics, mood, production techniques
+- Be extremely specific and detailed
+
+EXAMPLE OUTPUTS:
+
+"A cappella arrangement with pure vocal harmonies, no instrumental accompaniment. Features a lead soprano voice with rich layered backing vocals creating lush harmonies. 70 BPM, peaceful and soothing mood."
+
+"Intimate jazz duet featuring conversational interplay between a warm male baritone and a silky female alto voice. Accompanied by brushed jazz drums, walking upright bass, and gentle piano comping. 95 BPM, late-night jazz club atmosphere."
+
+"High-energy K-Pop dance track with a bright, clear female vocal. Catchy hook melody, driving beat with punchy kicks and intricate hi-hat programming. Layered synth hooks and EDM-style buildups. 128 BPM, confident energy."
+
+Output ONLY the final prompt paragraph, nothing else."""
+}
+
+
+def call_groq(api_key: str, system: str, user_prompt: str, context: str = "") -> str:
+    """Groq API 호출 - 강화된 에러 핸들링"""
+    try:
+        client = Groq(api_key=api_key)
+        
+        messages = [{"role": "system", "content": system}]
+        if context:
+            messages.append({"role": "user", "content": f"Previous work:\n{context}\n\nTask: {user_prompt}"})
+        else:
+            messages.append({"role": "user", "content": user_prompt})
+        
+        completion = client.chat.completions.create(
+            model="llama-3.3-70b-versatile",
+            messages=messages,
+            temperature=0.9,
+            max_completion_tokens=4096,
+            top_p=1,
+            stream=False
+        )
+        
+        if completion is None:
+            return "Error: API 응답이 없습니다."
+        if not hasattr(completion, 'choices') or not completion.choices:
+            return "Error: API 응답에 choices가 없습니다."
+        if completion.choices[0].message is None:
+            return "Error: API 응답에 message가 없습니다."
+        if completion.choices[0].message.content is None:
+            return "Error: API 응답 content가 비��있습니다."
+            
+        return completion.choices[0].message.content
+        
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+
+def clean_lyrics(text: str) -> str:
+    """가사 후처리 - HeartMuLa 포맷 최적화"""
+    import re
+    
+    if text is None:
+        return ""
+    if not isinstance(text, str):
+        return str(text)
+    if not text.strip():
+        return ""
+    
+    lines = text.split('\n')
+    cleaned_lines = []
+    
+    for line in lines:
+        if not line.strip():
+            cleaned_lines.append('')
+            continue
+        
+        skip_patterns = [
+            r'^\s*\*',
+            r'^\s*>',
+            r'^\s*---',
+            r'^\s*###',
+            r'^\s*\*\*.*\*\*\s*$',
+            r'^\(\s*.*\s*\)$',
+            r'^\s*–\s*\*',
+        ]
+        
+        skip = False
+        for pattern in skip_patterns:
+            if re.match(pattern, line):
+                skip = True
+                break
+        
+        if skip:
+            continue
+        
+        line = re.sub(r'\s*\([A-Za-z].*?\)\s*$', '', line)
+        line = re.sub(r'\*\*(.*?)\*\*', r'\1', line)
+        
+        # HeartMuLa 태그 정규화
+        line = re.sub(r'\[Pre.?Chorus\]', '[Prechorus]', line, flags=re.IGNORECASE)
+        line = re.sub(r'\[Post.?Chorus\]', '[Chorus]', line, flags=re.IGNORECASE)  # HeartMuLa는 PostChorus 없음
+        line = re.sub(r'\[Build.?Up\]', '[Prechorus]', line, flags=re.IGNORECASE)
+        line = re.sub(r'\[Break\]', '[Interlude]', line, flags=re.IGNORECASE)
+        line = re.sub(r'\[Transition\]', '[Interlude]', line, flags=re.IGNORECASE)
+        
+        if line.strip():
+            cleaned_lines.append(line.strip())
+    
+    result = '\n'.join(cleaned_lines)
+    result = re.sub(r'\n{3,}', '\n\n', result)
+    
+    return result.strip()
+
+
+def clean_tags(tags: str) -> str:
+    """태그 정리 - HeartMuLa 포맷 (콤마 구분, 공백 없음)"""
+    if not tags:
+        return ""
+    
+    # 공백 제거, 소문자 변환
+    tags = tags.lower().strip()
+    
+    # 다양한 구분자를 콤마로 통일
+    tags = tags.replace(', ', ',').replace(' ,', ',').replace('  ', ' ')
+    tags = tags.replace(' ', ',').replace(',,', ',')
+    
+    # 앞뒤 콤마 제거
+    tags = tags.strip(',')
+    
+    # 중복 제거
+    tag_list = [t.strip() for t in tags.split(',') if t.strip()]
+    unique_tags = list(dict.fromkeys(tag_list))
+    
+    return ','.join(unique_tags)
+
+
+def generate_lyrics_soma(
+    api_key: str, theme: str, genre: str, mood: str, 
+    language: str, vocal_type: str, additional: str, progress=gr.Progress()
+):
+    """SOMA 가사 생성 - HeartMuLa 최적화"""
+    if not api_key or not api_key.strip():
+        return "❌ Groq API Key 필요", "", "", "", ""
+    if not theme or not theme.strip():
+        return "❌ 주제를 입력하세요", "", "", "", ""
+    
+    base_prompt = f"""Create PROFESSIONAL lyrics optimized for HeartMuLa/MiniMax Music:
+- Theme: {theme}
+- Genre: {genre}
+- Mood: {mood}
+- Language: {language}
+- Vocal Type: {vocal_type}
+{f'- Additional: {additional}' if additional else ''}
+
+CRITICAL - USE HeartMuLa STRUCTURE TAGS:
+Available: [Intro], [Verse], [Prechorus], [Chorus], [Bridge], [Interlude], [Hook], [Outro], [Inst], [Solo]
+
+REQUIRED STRUCTURE (minimum):
+1. [Intro] - Set the mood
+2. [Verse] x2-3 - Tell the story
+3. [Prechorus] - Build tension (note: Prechorus, not Pre Chorus)
+4. [Chorus] x2-3 - Main hook (MOST important!)
+5. [Bridge] - Emotional contrast
+6. [Outro] - Conclusion
+
+{"For A Cappella: Include harmony sections with 'ooh', 'aah', humming." if "cappella" in vocal_type.lower() else ""}
+{"For Duet: Mark vocal exchanges clearly." if "duet" in vocal_type.lower() else ""}
+{"For Group/Choir: Include anthemic chants, call-and-response patterns." if "group" in vocal_type.lower() or "choir" in vocal_type.lower() else ""}
+
+Create lyrics with PERFECT HeartMuLa structure!"""
+
+    try:
+        progress(0.2, desc="🎤 작사가 - 초안 작성...")
+        draft = call_groq(api_key, LYRICS_AGENTS["lyricist"], base_prompt)
+        if draft.startswith("Error:"):
+            return f"❌ 작사가 오류: {draft}", "", "", "", ""
+        
+        progress(0.4, desc="🎹 프로듀서 - 구조 최적화...")
+        structured = call_groq(api_key, LYRICS_AGENTS["producer"], 
+                              f"Optimize structure for {genre} {vocal_type}. Use HeartMuLa tags.", draft)
+        if structured.startswith("Error:"):
+            return f"❌ 프로듀서 오류: {structured}", draft, "", "", ""
+        
+        progress(0.6, desc="💫 감성 디렉터 - 감정 강화...")
+        emotional = call_groq(api_key, LYRICS_AGENTS["emotion_director"],
+                             f"Enhance emotional impact for {mood}.", structured)
+        if emotional.startswith("Error:"):
+            return f"❌ 감성 디렉터 오류: {emotional}", draft, structured, "", ""
+        
+        progress(0.8, desc="✨ 최종 ���집 - 품질 검증...")
+        final = call_groq(api_key, LYRICS_AGENTS["final_editor"],
+                         "Output ONLY clean lyrics with HeartMuLa tags. No translations, no markdown.", emotional)
+        if final.startswith("Error:"):
+            return f"❌ 최종 편집 오류: {final}", draft, structured, emotional, ""
+        
+        final_cleaned = clean_lyrics(final)
+        
+        progress(1.0, desc="✅ 완료!")
+        return "✅ 가사 생성 완료!", draft, structured, emotional, final_cleaned
+        
+    except Exception as e:
+        return f"❌ 예외 발생: {str(e)}", "", "", "", ""
 
-            gr.Markdown(
-                """
-                ### Tips for Better Results
-                - Use structured lyrics with section tags
-                - Be specific with your style tags
-                - Try different temperature values for variety
-                - Shorter durations generate faster
-
-                ### Example Tags
-                - **Instruments:** piano, guitar, drums, synthesizer, violin, bass
-                - **Mood:** happy, sad, romantic, energetic, calm, melancholic
-                - **Genre:** pop, rock, jazz, classical, electronic, folk
-                - **Tempo:** fast, slow, upbeat, relaxed
-                """
+
+def quick_lyrics(api_key: str, theme: str, genre: str, mood: str, language: str, vocal_type: str, additional: str):
+    """빠른 가사 생성 - HeartMuLa 최적화"""
+    if not api_key or not api_key.strip():
+        return "❌ API Key가 필요합니다"
+    if not theme or not theme.strip():
+        return "❌ 주제를 입력하세요"
+    
+    prompt = f"""Create PROFESSIONAL song lyrics for HeartMuLa/MiniMax Music:
+- Theme: {theme}
+- Genre: {genre}
+- Mood: {mood}
+- Language: {language}
+- Vocal: {vocal_type}
+{f'- Special: {additional}' if additional else ''}
+
+USE HeartMuLa STRUCTURE (minimum 8-10 sections):
+[Intro] → [Verse] → [Prechorus] → [Chorus] → [Verse] → [Prechorus] → [Chorus] → [Bridge] → [Chorus] → [Outro]
+
+Available tags: [Intro], [Verse], [Prechorus], [Chorus], [Bridge], [Interlude], [Hook], [Outro], [Inst], [Solo]
+
+REQUIREMENTS:
+- [Chorus] must appear AT LEAST 2-3 times
+- Use [Prechorus] (not Pre Chorus) before each [Chorus]
+- Add [Bridge] before final [Chorus]
+
+OUTPUT ONLY: Structure tags + lyrics. NO translations, NO explanations."""
+
+    try:
+        result = call_groq(api_key, f"""You are a professional songwriter for HeartMuLa Music.
+Create lyrics with PERFECT HeartMuLa structure tag placement.
+{HEARTMULA_LYRICS_STRUCTURE}
+Output ONLY clean lyrics with optimal tags.""", prompt)
+        
+        if result.startswith("Error:"):
+            return f"❌ 가사 생성 실패: {result}"
+        
+        return clean_lyrics(result)
+    except Exception as e:
+        return f"❌ 예외 발생: {str(e)}"
+
+
+def generate_tags(api_key: str, genre: str, mood: str, instruments: str, tempo: str):
+    """HeartMuLa 스타일 태그 생성"""
+    if not api_key or not api_key.strip():
+        return "piano,happy,pop"
+    
+    prompt = f"""Generate HeartMuLa style tags:
+- Genre: {genre}
+- Mood: {mood}
+- Instruments: {instruments}
+- Tempo: {tempo}
+
+OUTPUT FORMAT: comma-separated tags WITHOUT spaces
+Example: piano,synthesizer,happy,pop,upbeat,romantic
+
+Generate 5-8 relevant tags:"""
+
+    try:
+        result = call_groq(api_key, PROMPT_AGENTS["tag_generator"], prompt)
+        if result.startswith("Error:"):
+            return "piano,happy,pop"
+        return clean_tags(result)
+    except:
+        return "piano,happy,pop"
+
+
+def augment_prompt_soma(
+    api_key: str, user_prompt: str, genre: str, mood: str,
+    tempo: str, vocal_type: str, instruments: str, reference_style: str, progress=gr.Progress()
+):
+    """SOMA 프롬프트 증강 (HeartMuLa 최적화)"""
+    if not api_key or not api_key.strip():
+        return "❌ Groq API Key 필요", ""
+    if not user_prompt or not user_prompt.strip():
+        return "❌ 기본 프롬프트를 입력하세요", ""
+    
+    base_info = f"""User's base idea: {user_prompt}
+Genre: {genre}
+Mood: {mood}
+Tempo: {tempo}
+Vocal Type: {vocal_type}
+Instruments: {instruments}
+Reference Style: {reference_style}"""
+
+    try:
+        progress(0.2, desc="🎸 장르 분석중...")
+        genre_analysis = call_groq(api_key, PROMPT_AGENTS["genre_specialist"], base_info)
+        if genre_analysis.startswith("Error:"):
+            return f"❌ 장르 분석 실패: {genre_analysis}", ""
+        
+        progress(0.4, desc="🎛️ 사운드 설계중...")
+        sound_design = call_groq(api_key, PROMPT_AGENTS["sound_designer"], 
+                                f"Design sounds for:\n{base_info}", genre_analysis)
+        if sound_design.startswith("Error:"):
+            return f"❌ 사운드 설계 실패: {sound_design}", ""
+        
+        progress(0.55, desc="🎤 보컬 설정중...")
+        vocal_design = call_groq(api_key, PROMPT_AGENTS["vocal_director"],
+                                f"Define vocals for:\n{base_info}", sound_design)
+        if vocal_design.startswith("Error:"):
+            return f"❌ 보컬 설정 실패: {vocal_design}", ""
+        
+        progress(0.7, desc="🏷️ 태그 생성중...")
+        tags = call_groq(api_key, PROMPT_AGENTS["tag_generator"],
+                        f"Generate tags for: {genre}, {mood}, {instruments}, {tempo}")
+        tags_cleaned = clean_tags(tags) if not tags.startswith("Error:") else "piano,happy,pop"
+        
+        progress(0.85, desc="✨ 프롬프트 생성중...")
+        final_prompt = call_groq(
+            api_key, 
+            PROMPT_AGENTS["prompt_synthesizer"],
+            f"""Synthesize into ONE music production prompt (150-200 words):
+Base: {user_prompt}
+Genre Analysis: {genre_analysis}
+Sound Design: {sound_design}
+Vocal Design: {vocal_design}
+Reference Style: {reference_style}
+
+Output ONLY the final prompt paragraph in English."""
+        )
+        
+        if final_prompt.startswith("Error:"):
+            return f"❌ 프롬프트 합성 실패: {final_prompt}", ""
+        
+        progress(1.0, desc="✅ 완료!")
+        return final_prompt.strip(), tags_cleaned
+        
+    except Exception as e:
+        return f"❌ 예외 발생: {str(e)}", ""
+
+
+def generate_music(api_key: str, model: str, prompt: str, lyrics: str,
+                   sample_rate: int, bitrate: int, audio_format: str):
+    """MiniMax 음악 생성 (모델 2.5)"""
+    if not api_key or not api_key.strip():
+        return None, "❌ MiniMax API Key 필요", ""
+    if not prompt or not prompt.strip():
+        return None, "❌ 프롬프트 필요", ""
+    
+    url = "https://api.minimax.io/v1/music_generation"
+    headers = {"Content-Type": "application/json", "Authorization": f"Bearer {api_key}"}
+    
+    payload = {
+        "model": model,
+        "prompt": prompt,
+        "audio_setting": {
+            "sample_rate": sample_rate, 
+            "bitrate": bitrate, 
+            "format": audio_format
+        }
+    }
+    if lyrics and lyrics.strip():
+        payload["lyrics"] = lyrics
+    
+    try:
+        response = requests.post(url, headers=headers, json=payload, timeout=600)
+        
+        if response is None:
+            return None, "❌ API 응답이 없습니다.", ""
+        
+        try:
+            result = response.json()
+        except Exception as json_err:
+            return None, f"❌ JSON 파싱 실패: {str(json_err)}", response.text[:500] if response.text else ""
+        
+        if result is None:
+            return None, "❌ API 응답이 비어있습니다.", ""
+        
+        # JSON 출력용 (audio hex 데이터 축약)
+        result_for_display = result.copy()
+        if "data" in result_for_display and isinstance(result_for_display["data"], dict):
+            data_copy = result_for_display["data"].copy()
+            if "audio" in data_copy and isinstance(data_copy["audio"], str) and len(data_copy["audio"]) > 100:
+                data_copy["audio"] = f"[HEX DATA - {len(data_copy['audio'])} chars]"
+            result_for_display["data"] = data_copy
+        
+        json_output = json.dumps(result_for_display, ensure_ascii=False, indent=2)
+        
+        base_resp = result.get("base_resp", {})
+        status_code = base_resp.get("status_code", -1)
+        status_msg = base_resp.get("status_msg", "")
+        
+        if status_code != 0:
+            return None, f"❌ API 오류: {status_msg} (code: {status_code})", json_output
+        
+        data = result.get("data", {})
+        if not data:
+            return None, "❌ 응답에 data가 없습니다.", json_output
+        
+        audio_hex = data.get("audio")
+        audio_status = data.get("status")
+        
+        if audio_status != 2:
+            return None, f"⏳ 생성 중... (status: {audio_status})", json_output
+        
+        if not audio_hex:
+            return None, "❌ 응답에 audio 데이터가 없습니다.", json_output
+        
+        try:
+            import tempfile
+            import time
+            
+            audio_bytes = bytes.fromhex(audio_hex)
+            file_ext = audio_format if audio_format else "mp3"
+            timestamp = int(time.time())
+            filename = f"minimax_music_{timestamp}.{file_ext}"
+            
+            save_paths = [
+                os.path.join(tempfile.gettempdir(), filename),
+                os.path.join(os.getcwd(), filename),
+                filename
+            ]
+            
+            saved_path = None
+            for path in save_paths:
+                try:
+                    with open(path, "wb") as f:
+                        f.write(audio_bytes)
+                    saved_path = path
+                    break
+                except Exception:
+                    continue
+            
+            if not saved_path:
+                return None, "❌ 파일 저장 실패", json_output
+            
+            extra_info = result.get("extra_info", {})
+            duration_ms = extra_info.get("music_duration", 0)
+            duration_sec = duration_ms / 1000 if duration_ms else 0
+            file_size_kb = len(audio_bytes) / 1024
+            
+            return saved_path, f"✅ 음악 생성 완료! ({duration_sec:.1f}초, {file_size_kb:.0f}KB)", json_output
+            
+        except ValueError as hex_err:
+            return None, f"❌ HEX 디코딩 실패: {str(hex_err)}", json_output
+        except Exception as save_err:
+            return None, f"❌ 파일 저장 실패: {str(save_err)}", json_output
+            
+    except requests.exceptions.Timeout:
+        return None, "❌ 요청 시간 초과 (10분)", ""
+    except requests.exceptions.ConnectionError:
+        return None, "❌ 연결 오류", ""
+    except Exception as e:
+        return None, f"❌ 예외 발생: {str(e)}", ""
+
+
+def load_example_from_dropdown(selection):
+    """Dropdown 선택시 예제 프롬프트 및 태그 로드"""
+    if not selection:
+        return "", ""
+    
+    mapping = {
+        "🎤 A Cappella - 순수 보컬 하모니": "🎤 A Cappella (아카펠라)",
+        "👥 Group Harmony - 파워풀 합창": "👥 Group Harmony (그룹 하모니)",
+        "🎷 Jazz Duet - 남녀 듀엣": "🎷 Jazz Duet (재즈 듀엣)",
+        "🎸 Multi-Style - 스타일 전환": "🎸 Multi-Style (멀티스타일)",
+        "🌃 Urban Chill - R&B": "🌃 Urban Chill (어반 칠)",
+        "🎹 Jazz Club - 라이브": "🎹 Jazz Club (재즈 클럽)",
+        "🪩 Retro Disco - 80년대": "🪩 Retro Disco (레트로 디스코)",
+        "🎬 Film Score - 시네마틱": "🎬 Film Score (영화 스코어)",
+        "🎵 K-Pop Dance - 고에너지": "🎵 K-Pop Dance (K-Pop 댄스)",
+        "🎻 Orchestral Ballad - 웅장": "🎻 Orchestral Ballad (오케스트라 발라드)",
+        "🔥 HeartMuLa Default - 기본": "🔥 HeartMuLa Default (기본 예시)"
+    }
+    
+    key = mapping.get(selection)
+    if key and key in EXAMPLE_PROMPTS:
+        return EXAMPLE_PROMPTS[key]["prompt"], EXAMPLE_PROMPTS[key]["tags"]
+    return "", ""
+
+
+# ============================================================
+# 🎨 Comic Classic Theme - Toon Playground (Document 2 기반)
+# ============================================================
+
+css = """
+/* ===== 🎨 Google Fonts Import ===== */
+@import url('https://fonts.googleapis.com/css2?family=Bangers&family=Comic+Neue:wght@400;700&display=swap');
+
+/* ===== 🎨 Comic Classic 배경 - 빈티지 페이퍼 + 도트 패턴 ===== */
+.gradio-container {
+    background-color: #FEF9C3 !important;
+    background-image: 
+        radial-gradient(#1F2937 1px, transparent 1px) !important;
+    background-size: 20px 20px !important;
+    min-height: 100vh !important;
+    font-family: 'Comic Neue', cursive, sans-serif !important;
+}
+
+/* ===== 허깅페이스 상단 요소 숨김 ===== */
+.huggingface-space-header,
+#space-header,
+.space-header,
+[class*="space-header"],
+.svelte-1ed2p3z,
+.space-header-badge,
+.header-badge {
+    display: none !important;
+}
+
+/* ===== Footer 완전 숨김 ===== */
+footer,
+.footer,
+.gradio-container footer,
+.built-with {
+    display: none !important;
+}
+
+/* ===== 메인 컨테이너 ===== */
+#col-container { 
+    max-width: 1400px; 
+    margin: 0 auto; 
+}
+
+/* ===== 🎨 헤더 타이틀 - 코믹 스타일 ===== */
+.header-title h1 {
+    font-family: 'Bangers', cursive !important;
+    color: #1F2937 !important;
+    font-size: 3.2rem !important;
+    font-weight: 400 !important;
+    text-align: center !important;
+    margin-bottom: 0.5rem !important;
+    text-shadow: 
+        4px 4px 0px #FACC15,
+        6px 6px 0px #1F2937 !important;
+    letter-spacing: 3px !important;
+    -webkit-text-stroke: 2px #1F2937 !important;
+}
+
+/* ===== 🎨 서브타이틀 ===== */
+.subtitle-text {
+    text-align: center !important;
+    font-family: 'Comic Neue', cursive !important;
+    font-size: 1.1rem !important;
+    color: #1F2937 !important;
+    margin-bottom: 1.5rem !important;
+    font-weight: 700 !important;
+}
+
+/* ===== 🎨 섹션 타이틀 ===== */
+.section-title {
+    font-family: 'Bangers', cursive !important;
+    color: #1F2937 !important;
+    font-size: 1.8rem !important;
+    border-bottom: 4px solid #3B82F6 !important;
+    padding-bottom: 8px !important;
+    margin-bottom: 16px !important;
+    text-shadow: 2px 2px 0px #FACC15 !important;
+}
+
+/* ===== 🎨 카드/패널 - 만화 프레임 스타일 ===== */
+.gr-panel,
+.gr-box,
+.gr-form,
+.block,
+.gr-group {
+    background: #FFFFFF !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px !important;
+    box-shadow: 6px 6px 0px #1F2937 !important;
+    transition: all 0.2s ease !important;
+}
+
+.gr-panel:hover,
+.block:hover {
+    transform: translate(-2px, -2px) !important;
+    box-shadow: 8px 8px 0px #1F2937 !important;
+}
+
+/* ===== 🎨 입력 필드 (Textbox) ===== */
+textarea, 
+input[type="text"], 
+input[type="number"],
+input[type="password"] {
+    background: #FFFFFF !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px !important;
+    color: #1F2937 !important;
+    font-family: 'Comic Neue', cursive !important;
+    font-size: 1rem !important;
+    font-weight: 700 !important;
+    transition: all 0.2s ease !important;
+}
+
+textarea:focus, 
+input[type="text"]:focus, 
+input[type="number"]:focus,
+input[type="password"]:focus {
+    border-color: #3B82F6 !important;
+    box-shadow: 4px 4px 0px #3B82F6 !important;
+    outline: none !important;
+}
+
+textarea::placeholder {
+    color: #9CA3AF !important;
+    font-weight: 400 !important;
+}
+
+/* ===== 🎨 Primary 버튼 - 코믹 블루 ===== */
+.gr-button-primary,
+button.primary,
+.gr-button.primary {
+    background: #3B82F6 !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px !important;
+    color: #FFFFFF !important;
+    font-family: 'Bangers', cursive !important;
+    font-weight: 400 !important;
+    font-size: 1.2rem !important;
+    letter-spacing: 2px !important;
+    padding: 12px 24px !important;
+    box-shadow: 5px 5px 0px #1F2937 !important;
+    transition: all 0.1s ease !important;
+    text-shadow: 1px 1px 0px #1F2937 !important;
+}
+
+.gr-button-primary:hover,
+button.primary:hover,
+.gr-button.primary:hover {
+    background: #2563EB !important;
+    transform: translate(-2px, -2px) !important;
+    box-shadow: 7px 7px 0px #1F2937 !important;
+}
+
+.gr-button-primary:active,
+button.primary:active,
+.gr-button.primary:active {
+    transform: translate(3px, 3px) !important;
+    box-shadow: 2px 2px 0px #1F2937 !important;
+}
+
+/* ===== 🎨 Secondary 버튼 - 코믹 레드 ===== */
+.gr-button-secondary,
+button.secondary {
+    background: #EF4444 !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px !important;
+    color: #FFFFFF !important;
+    font-family: 'Bangers', cursive !important;
+    font-weight: 400 !important;
+    font-size: 1.1rem !important;
+    letter-spacing: 1px !important;
+    box-shadow: 4px 4px 0px #1F2937 !important;
+    transition: all 0.1s ease !important;
+    text-shadow: 1px 1px 0px #1F2937 !important;
+}
+
+.gr-button-secondary:hover,
+button.secondary:hover {
+    background: #DC2626 !important;
+    transform: translate(-2px, -2px) !important;
+    box-shadow: 6px 6px 0px #1F2937 !important;
+}
+
+/* ===== 🎨 Generate 버튼 - 코믹 그린 ===== */
+.generate-btn {
+    background: #10B981 !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px !important;
+    color: #FFFFFF !important;
+    font-family: 'Bangers', cursive !important;
+    font-weight: 400 !important;
+    font-size: 1.3rem !important;
+    letter-spacing: 2px !important;
+    box-shadow: 5px 5px 0px #1F2937 !important;
+    text-shadow: 1px 1px 0px #1F2937 !important;
+}
+
+.generate-btn:hover {
+    background: #059669 !important;
+    transform: translate(-2px, -2px) !important;
+    box-shadow: 7px 7px 0px #1F2937 !important;
+}
+
+/* ===== 🎨 아코디언 - 말풍선 스타일 ===== */
+.gr-accordion {
+    background: #FACC15 !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px !important;
+    box-shadow: 4px 4px 0px #1F2937 !important;
+}
+
+.gr-accordion-header {
+    color: #1F2937 !important;
+    font-family: 'Comic Neue', cursive !important;
+    font-weight: 700 !important;
+    font-size: 1.1rem !important;
+}
+
+/* ===== 🎨 Dropdown ===== */
+.gr-dropdown,
+select {
+    background: #FFFFFF !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px !important;
+    color: #1F2937 !important;
+    font-family: 'Comic Neue', cursive !important;
+    font-weight: 700 !important;
+}
+
+/* ===== 🎨 라벨 스타일 ===== */
+label,
+.gr-input-label,
+.gr-block-label {
+    color: #1F2937 !important;
+    font-family: 'Comic Neue', cursive !important;
+    font-weight: 700 !important;
+    font-size: 1rem !important;
+}
+
+/* ===== 🎨 오디오 플레이어 ===== */
+.gr-audio,
+audio {
+    border: 4px solid #1F2937 !important;
+    border-radius: 8px !important;
+    box-shadow: 6px 6px 0px #1F2937 !important;
+}
+
+/* ===== 🎨 코드 블록 ===== */
+.gr-code,
+pre,
+code {
+    background: #1F2937 !important;
+    color: #10B981 !important;
+    font-family: 'Courier New', monospace !important;
+    border: 3px solid #10B981 !important;
+    border-radius: 8px !important;
+    box-shadow: 4px 4px 0px #10B981 !important;
+}
+
+/* ===== 🎨 마크다운 ===== */
+.gr-markdown {
+    font-family: 'Comic Neue', cursive !important;
+    color: #1F2937 !important;
+}
+
+.gr-markdown h1, .gr-markdown h2, .gr-markdown h3 {
+    font-family: 'Bangers', cursive !important;
+    color: #1F2937 !important;
+    text-shadow: 2px 2px 0px #FACC15 !important;
+}
+
+/* ===== 🎨 탭 스타일 ===== */
+.gr-tab-nav {
+    background: #FACC15 !important;
+    border: 3px solid #1F2937 !important;
+    border-radius: 8px 8px 0 0 !important;
+}
+
+.gr-tab-nav button {
+    font-family: 'Comic Neue', cursive !important;
+    font-weight: 700 !important;
+    color: #1F2937 !important;
+}
+
+.gr-tab-nav button.selected {
+    background: #3B82F6 !important;
+    color: #FFFFFF !important;
+}
+
+/* ===== 🎨 상태 표시 ===== */
+.status-box textarea {
+    background: #1F2937 !important;
+    color: #10B981 !important;
+    font-family: 'Courier New', monospace !important;
+    font-size: 0.9rem !important;
+    border: 3px solid #10B981 !important;
+    border-radius: 8px !important;
+}
+
+/* ===== 🎨 스크롤바 - 코믹 스타일 ===== */
+::-webkit-scrollbar {
+    width: 12px;
+    height: 12px;
+}
+
+::-webkit-scrollbar-track {
+    background: #FEF9C3;
+    border: 2px solid #1F2937;
+}
+
+::-webkit-scrollbar-thumb {
+    background: #3B82F6;
+    border: 2px solid #1F2937;
+    border-radius: 0px;
+}
+
+::-webkit-scrollbar-thumb:hover {
+    background: #EF4444;
+}
+
+/* ===== 🎨 선택 하이라이트 ===== */
+::selection {
+    background: #FACC15;
+    color: #1F2937;
+}
+
+/* ===== 🎨 Row/Column 간격 ===== */
+.gr-row {
+    gap: 1.5rem !important;
+}
+
+.gr-column {
+    gap: 1rem !important;
+}
+
+/* ===== 🎨 태그 입력 특별 스타일 ===== */
+.tag-input textarea {
+    background: #FEF3C7 !important;
+    border: 3px dashed #F59E0B !important;
+    font-family: 'Courier New', monospace !important;
+}
+
+/* ===== 반응형 조정 ===== */
+@media (max-width: 768px) {
+    .header-title h1 {
+        font-size: 2rem !important;
+        text-shadow: 
+            3px 3px 0px #FACC15,
+            4px 4px 0px #1F2937 !important;
+    }
+    
+    .gr-button-primary,
+    button.primary {
+        padding: 10px 16px !important;
+        font-size: 1rem !important;
+    }
+}
+
+/* ===== 🎨 특수 효과 - 반짝임 ===== */
+@keyframes sparkle {
+    0%, 100% { opacity: 1; }
+    50% { opacity: 0.7; }
+}
+
+.sparkle {
+    animation: sparkle 2s ease-in-out infinite;
+}
+"""
+
+# ============================================================
+# Gradio UI
+# ============================================================
+
+with gr.Blocks(css=css, title="🎵 SOMA Music Studio", theme=gr.themes.Soft(primary_hue="indigo")) as demo:
+    
+    # HOME Badge
+    gr.HTML("""
+        <div style="text-align: center; margin: 20px 0 10px 0;">
+            <a href="https://huggingface.co/HeartMuLa" target="_blank" style="text-decoration: none;">
+                <img src="https://img.shields.io/badge/🎵_HeartMuLa-Official-ff6b6b?style=for-the-badge&labelColor=1F2937" alt="HeartMuLa">
+            </a>
+            <a href="https://www.minimax.io" target="_blank" style="text-decoration: none; margin-left: 10px;">
+                <img src="https://img.shields.io/badge/🎹_MiniMax-Music_2.5-3B82F6?style=for-the-badge&labelColor=1F2937" alt="MiniMax">
+            </a>
+        </div>
+    """)
+    
+    # Header
+    gr.Markdown("""
+    # 🎵 SOMA MUSIC STUDIO 🎶
+    """, elem_classes="header-title")
+    
+    gr.Markdown("""
+    <p class="subtitle-text">💫 HeartMuLa + MiniMax Music 2.5 + SOMA Multi-Agent = 최고 품질 AI 음악 생성 💫</p>
+    <p class="subtitle-text">🎤 A Cappella | 🎷 Jazz Duet | 🎸 Multi-Style | 🎬 Film Score | 🎵 K-Pop 지원</p>
+    """)
+    
+    # API Keys
+    GROQ_KEY = os.environ.get("GROQ_API_KEY", "")
+    MINIMAX_KEY = os.environ.get("MINIMAX_API_KEY", "")
+    
+    with gr.Accordion("🔑 API Keys", open=not (GROQ_KEY and MINIMAX_KEY)):
+        with gr.Row():
+            groq_key = gr.Textbox(
+                label="🦙 Groq API Key (가사/프롬프트 생성용)",
+                type="password",
+                value=GROQ_KEY,
+                placeholder="gsk_..." if not GROQ_KEY else "✅ Secret 로드됨",
+                interactive=not bool(GROQ_KEY)
+            )
+            minimax_key = gr.Textbox(
+                label="🎹 MiniMax API Key (음악 생성용)",
+                type="password",
+                value=MINIMAX_KEY,
+                placeholder="API Key" if not MINIMAX_KEY else "✅ Secret 로드됨",
+                interactive=not bool(MINIMAX_KEY)
+            )
+    
+    with gr.Row(equal_height=False):
+        # ========== 좌측: 가사 생성 ==========
+        with gr.Column(scale=1, min_width=400):
+            gr.Markdown("## 📝 LYRICS GENERATOR", elem_classes="section-title")
+            
+            theme_input = gr.Textbox(
+                label="🎯 노래 주제",
+                placeholder="예: 이별 후 성장, 꿈을 향한 도전, 사랑의 고백, 우정의 힘...",
+                lines=2
+            )
+            
+            with gr.Row():
+                lyrics_genre = gr.Dropdown(
+                    label="🎸 장르",
+                    choices=["K-Pop", "Pop", "R&B", "Hip-Hop", "Ballad", "Rock", 
+                            "EDM", "Trap", "Jazz", "Blues", "Folk", "Disco", "Cinematic"],
+                    value="K-Pop"
+                )
+                lyrics_mood = gr.Dropdown(
+                    label="💫 분위기",
+                    choices=["Empowering", "Melancholic", "Joyful", "Romantic",
+                            "Aggressive", "Dreamy", "Nostalgic", "Energetic", "Dark", 
+                            "Peaceful", "Confident", "Intimate"],
+                    value="Empowering"
+                )
+            
+            with gr.Row():
+                lyrics_language = gr.Dropdown(
+                    label="🌍 언어",
+                    choices=["English", "Korean", "Korean + English", "Japanese"],
+                    value="Korean"
+                )
+                lyrics_vocal_type = gr.Dropdown(
+                    label="🎤 보컬 타입",
+                    choices=["Solo Female", "Solo Male", "Female Duet", "Male Duet", 
+                            "Male-Female Duet", "A Cappella", "Group/Choir"],
+                    value="Group/Choir"
+                )
+            
+            lyrics_additional = gr.Textbox(
+                label="✨ 추가 지시 (선택)",
+                placeholder="특별 요청: 후렴구 강조, 랩 파트 추가, 특정 단어 포함...",
+                lines=1
+            )
+            
+            with gr.Row():
+                quick_btn = gr.Button("⚡ QUICK GENERATE", variant="secondary")
+                soma_lyrics_btn = gr.Button("🧠 SOMA GENERATE", variant="primary")
+            
+            lyrics_status = gr.Textbox(label="📊 상태", interactive=False, max_lines=1, elem_classes="status-box")
+            
+            with gr.Accordion("🔍 SOMA 작업 과정", open=False):
+                with gr.Row():
+                    with gr.Column():
+                        step1_out = gr.Textbox(label="1️⃣ 작사가", lines=4, interactive=False)
+                        step2_out = gr.Textbox(label="2️⃣ 프로듀서", lines=4, interactive=False)
+                    with gr.Column():
+                        step3_out = gr.Textbox(label="3️⃣ 감성 디렉터", lines=4, interactive=False)
+                        step4_out = gr.Textbox(label="4️⃣ 최종 편집", lines=4, interactive=False)
+            
+            final_lyrics = gr.Textbox(
+                label="✏️ 최종 가사 (편집 가능)",
+                lines=14,
+                placeholder="생성된 가사가 여기 표시됩니다...\n\n[Intro]\n\n[Verse]\n...\n\n[Chorus]\n..."
             )
+            
+            with gr.Accordion("📋 HeartMuLa 구조 태그 가이드", open=False):
+                gr.Markdown("""
+**HeartMuLa 태그:** `[Intro]` `[Verse]` `[Prechorus]` `[Chorus]` `[Bridge]` `[Interlude]` `[Hook]` `[Outro]` `[Inst]` `[Solo]`
 
-    generate_btn.click(
-        fn=generate_music,
-        inputs=[
-            lyrics_input,
-            tags_input,
-            max_duration,
-            temperature,
-            topk,
-            cfg_scale,
-        ],
-        outputs=audio_output,
-    )
+⚠️ **주의:** `[Pre Chorus]`가 아닌 `[Prechorus]` 사용 (공백 없음)
 
-    gr.Markdown(
-        """
-        ---
-        **Model:** [HeartMuLa-oss-3B](https://huggingface.co/HeartMuLa/HeartMuLa-oss-3B) |
-        **Paper:** [arXiv](https://arxiv.org/abs/2601.10547) |
-        **Code:** [GitHub](https://github.com/HeartMuLa/heartlib)
+**최적 구조 예시:**
+```
+[Intro] → [Verse] → [Prechorus] → [Chorus]
+→ [Verse] → [Prechorus] → [Chorus] 
+→ [Bridge] → [Chorus] → [Outro]
+```
 
-        *Licensed under Apache 2.0*
-        """
-    )
+**태그별 역할:**
+- `[Chorus]` - 가장 중요! 2-3회 반복, 기억에 남는 훅
+- `[Prechorus]` - 코러스 전 텐션 빌드업
+- `[Bridge]` - 감정 전환점, 최종 코러스 전 배치
+                """)
+        
+        # ========== 우측: 음악 생성 ==========
+        with gr.Column(scale=1, min_width=400):
+            gr.Markdown("## 🎵 MUSIC GENERATOR", elem_classes="section-title")
+            
+            # 예제 프롬프트 선택
+            gr.Markdown("### 📚 예제 프롬프트 (클릭하면 자동 입력)")
+            
+            example_dropdown = gr.Dropdown(
+                label="🎯 예제 선택",
+                choices=[
+                    "🎤 A Cappella - 순수 보컬 하모니",
+                    "👥 Group Harmony - 파워풀 합창",
+                    "🎷 Jazz Duet - 남녀 듀엣",
+                    "🎸 Multi-Style - 스타일 전환",
+                    "🌃 Urban Chill - R&B",
+                    "🎹 Jazz Club - 라이브",
+                    "🪩 Retro Disco - 80년대",
+                    "🎬 Film Score - 시네마틱",
+                    "🎵 K-Pop Dance - 고에너지",
+                    "🎻 Orchestral Ballad - 웅장",
+                    "🔥 HeartMuLa Default - 기본"
+                ],
+                value=None,
+                interactive=True
+            )
+            
+            gr.Markdown("### 🎛️ 프롬프트 설정")
+            
+            base_prompt = gr.Textbox(
+                label="💡 기본 아이디어",
+                placeholder="원하는 음악 스타일을 간단히 설명하세요...",
+                lines=2
+            )
+            
+            with gr.Row():
+                music_genre = gr.Dropdown(
+                    label="🎸 장르",
+                    choices=["K-Pop", "Pop", "R&B/Soul", "Hip-Hop/Trap", "EDM/House",
+                            "Rock", "Ballad", "Jazz", "Blues", "Lo-Fi", "Disco", 
+                            "Cinematic", "Classical"],
+                    value="K-Pop"
+                )
+                music_mood = gr.Dropdown(
+                    label="💫 분위기",
+                    choices=["Energetic", "Chill", "Emotional", "Dark", "Uplifting",
+                            "Romantic", "Aggressive", "Dreamy", "Confident", "Peaceful",
+                            "Nostalgic", "Epic"],
+                    value="Energetic"
+                )
+            
+            with gr.Row():
+                music_tempo = gr.Dropdown(
+                    label="⏱️ 템포",
+                    choices=["Very Slow (50-70 BPM)", "Slow (70-90 BPM)", "Medium (90-110 BPM)", 
+                            "Fast (110-130 BPM)", "Very Fast (130-150 BPM)"],
+                    value="Medium (90-110 BPM)"
+                )
+                music_vocal = gr.Dropdown(
+                    label="🎤 보컬",
+                    choices=["Female (Clear)", "Female (Warm)", "Female (Powerful)",
+                            "Male (Smooth)", "Male (Deep)", "Male (Raspy)",
+                            "Male-Female Duet", "A Cappella", "Group/Choir", 
+                            "Instrumental"],
+                    value="Group/Choir"
+                )
+            
+            music_instruments = gr.Textbox(
+                label="🎹 악기/사운드",
+                placeholder="예: piano, synthesizer, drums, bass, strings, saxophone...",
+                value="piano,synthesizer,drums,bass,strings"
+            )
+            
+            music_reference = gr.Dropdown(
+                label="🎯 레퍼런스 스타일",
+                choices=["None", "A Cappella", "Jazz Club", "Urban R&B", 
+                        "Retro Disco", "Film Score", "K-Pop Dance", "Orchestral Ballad"],
+                value="None"
+            )
+            
+            augment_btn = gr.Button("🚀 SOMA AUGMENT", variant="primary", size="lg")
+            
+            augmented_prompt = gr.Textbox(
+                label="✨ 증강된 프롬프트 (편집 가능)",
+                lines=5,
+                placeholder="SOMA가 생성한 고품질 프롬프트..."
+            )
+            
+            style_tags = gr.Textbox(
+                label="🏷️ 스타일 태그 (HeartMuLa 포맷: 콤마 구분, 공백 없음)",
+                placeholder="piano,happy,pop,upbeat,romantic",
+                lines=1,
+                elem_classes="tag-input"
+            )
+            
+            with gr.Accordion("⚙️ 생성 설정", open=False):
+                model_select = gr.Dropdown(
+                    label="🤖 모델", choices=["music-2.5"], value="music-2.5"
+                )
+                with gr.Row():
+                    sample_rate = gr.Dropdown(label="Sample Rate", choices=[44100], value=44100)
+                    bitrate = gr.Dropdown(label="Bitrate", choices=[128000, 192000, 256000], value=256000)
+                    audio_format = gr.Dropdown(label="Format", choices=["mp3", "wav"], value="mp3")
+            
+            generate_music_btn = gr.Button("🎶 GENERATE MUSIC!", variant="primary", size="lg", elem_classes="generate-btn")
+            
+            music_status = gr.Textbox(label="📊 상태", interactive=False, max_lines=1, elem_classes="status-box")
+            music_output = gr.Audio(label="🎧 생성된 음악", type="filepath")
+            
+            with gr.Accordion("📋 API 응답", open=False):
+                json_output = gr.Code(label="JSON Response", language="json", lines=6)
+    
+    # ========== 하단: 가이드 ==========
+    with gr.Accordion("📖 HeartMuLa & MiniMax 가이드", open=False):
+        gr.Markdown(f"""
+{HEARTMULA_TAG_GUIDE}
+
+---
 
+{HEARTMULA_LYRICS_STRUCTURE}
+        """)
+    
+    # ========== Event Handlers ==========
+    
+    # 예제 Dropdown 선택
+    example_dropdown.change(
+        fn=load_example_from_dropdown,
+        inputs=[example_dropdown],
+        outputs=[augmented_prompt, style_tags]
+    )
+    
+    # 빠른 가사 생성
+    quick_btn.click(
+        fn=quick_lyrics,
+        inputs=[groq_key, theme_input, lyrics_genre, lyrics_mood, lyrics_language, lyrics_vocal_type, lyrics_additional],
+        outputs=[final_lyrics]
+    )
+    
+    # SOMA 가사 생성
+    soma_lyrics_btn.click(
+        fn=generate_lyrics_soma,
+        inputs=[groq_key, theme_input, lyrics_genre, lyrics_mood, lyrics_language, lyrics_vocal_type, lyrics_additional],
+        outputs=[lyrics_status, step1_out, step2_out, step3_out, step4_out]
+    ).then(
+        fn=lambda x: x,
+        inputs=[step4_out],
+        outputs=[final_lyrics]
+    )
+    
+    # SOMA 프롬프트 증강
+    augment_btn.click(
+        fn=augment_prompt_soma,
+        inputs=[groq_key, base_prompt, music_genre, music_mood, music_tempo, music_vocal, music_instruments, music_reference],
+        outputs=[augmented_prompt, style_tags]
+    )
+    
+    # 음악 생성
+    generate_music_btn.click(
+        fn=generate_music,
+        inputs=[minimax_key, model_select, augmented_prompt, final_lyrics, sample_rate, bitrate, audio_format],
+        outputs=[music_output, music_status, json_output]
+    )
 
 
-demo.launch()
+if __name__ == "__main__":
+    demo.launch()