ComfyUI-Qwen3TTS-Emotional

🎭 Qwen3-TTS Emotional Voice Clone (Ultimate Edition)

Advanced voice cloning with 80+ emotion presets and multi-emotion mixing for ComfyUI.

🌟 Features

80+ Emotion Presets

Transform your voice clones with comprehensive emotional range:

• Core Emotions: neutral, angry, sad, happy, fearful, disgusted, surprised
• Intensity Variations: annoyed → angry → furious → enraged
• Social/Interpersonal: sarcastic, mocking, condescending, dismissive, sympathetic, apologetic, pleading
• Confidence Spectrum: confident, arrogant, smug, insecure, hesitant, timid, defeated, hopeless
• Energy Levels: excited, enthusiastic, energetic, bored, tired, exhausted, lethargic, calm
• Stress/Tension: nervous, panicked, frantic, stressed, overwhelmed, relaxed, serene
• Attitude/Mood: playful, teasing, flirtatious, seductive, mysterious, ominous, menacing, threatening
• Physical States: drunk, breathless, whispering, shouting, in_pain, sick
• Complex Emotions: bitter, manic, reluctant, desperate, jealous, frustrated, relieved, guilty, ashamed, proud
• Character Presets: rick_like, jerry_like, morty_like (Rick & Morty inspired)

Multi-Emotion Mixing

• Primary Emotion: Main emotional tone (required)
• Secondary Emotion: Blend with primary for nuanced delivery (optional)
• Tertiary Emotion: Add third layer of complexity (optional)
• Intensity Control: 0.0x to 2.0x multiplier for emotion strength

Voice Cloning Modes

• Fast Mode (x_vector_only=True): Quick generation, speaker embedding only
• Accurate Mode (x_vector_only=False): High-quality cloning with reference transcript

📋 Requirements

• ComfyUI
• ComfyUI-Qwen-TTS (base package)
• Python 3.10+
• CUDA-capable GPU (recommended) or CPU

Hardware Compatibility: Works on any system that can run Qwen3-TTS (NVIDIA GPUs, AMD GPUs with ROCm, Mac M1/M2, CPU)

🚀 Installation

Method 1: ComfyUI Manager (Recommended - Coming Soon)

1. Open ComfyUI Manager
2. Search for "Qwen3-TTS Emotional"
3. Click Install
4. Restart ComfyUI

Method 2: Manual Installation

Step 1: Install the base Qwen3-TTS package first

cd ComfyUI/custom_nodes
git clone https://github.com/flybirdxx/ComfyUI-Qwen-TTS.git
cd ComfyUI-Qwen-TTS
pip install -r requirements.txt

Step 2: Add the Emotional Voice Clone node

# Download qwen3_emotional_clone.py from this repo
# Place it in: ComfyUI/custom_nodes/ComfyUI-Qwen-TTS/

Step 3: Update __init__.py

Add these lines to ComfyUI/custom_nodes/ComfyUI-Qwen-TTS/__init__.py:

# Import emotional clone nodes
try:
    from .qwen3_emotional_clone import (
        NODE_CLASS_MAPPINGS as EMOTIONAL_MAPPINGS,
        NODE_DISPLAY_NAME_MAPPINGS as EMOTIONAL_DISPLAY
    )
    EMOTIONAL_LOADED = True
except Exception as e:
    print(f"⚠️ Failed to load emotional clone nodes: {e}")
    EMOTIONAL_LOADED = False

# Add to the existing NODE_CLASS_MAPPINGS section
if EMOTIONAL_LOADED:
    NODE_CLASS_MAPPINGS.update(EMOTIONAL_MAPPINGS)
    NODE_DISPLAY_NAME_MAPPINGS.update(EMOTIONAL_DISPLAY)
    print("✅ Emotional Voice Clone node loaded!")

Step 4: Restart ComfyUI

💡 Usage

Basic Usage

1. Add the node: Right-click → Qwen3-TTS → 🎭 Qwen3-TTS Emotional Voice Clone (Ultimate)
2. Connect reference audio: Load your voice sample (5-30 seconds recommended)
3. Enter text: What you want the voice to say
4. Select emotion: Choose from 80+ presets
5. Generate: Queue the prompt

Single Emotion Example

Primary Emotion: sad
Secondary Emotion: none
Tertiary Emotion: none
Intensity: 1.0x
Result: Pure sad delivery

Multi-Emotion Mixing Examples

Bitter (Pre-mixed)

Primary: sad
Secondary: angry
Intensity: 1.0x
= Bitter, resentful tone

Manic Energy

Primary: excited
Secondary: nervous  
Tertiary: breathless
Intensity: 1.5x
= Frantic, overwhelming energy

Maximum Confidence

Primary: confident
Secondary: arrogant
Tertiary: dismissive
Intensity: 2.0x
= Peak smugness and authority

Defeated Character

Primary: defeated
Secondary: apologetic
Tertiary: whispering
Intensity: 1.2x
= Complete emotional breakdown

Advanced Settings

Fast Generation (3-5 seconds):

• x_vector_only: True (enabled)
• ref_audio_transcript: Leave empty
• Good for testing, quick iterations

High Quality (15-20 seconds):

• x_vector_only: False (disabled)
• ref_audio_transcript: Fill in what the reference audio says
• Better voice matching, complete sentences

Controlling Length:

• Use randomize seed to try different generation lengths
• When you get a good long generation, note the seed
• Use that fixed seed for consistent results

🎨 Emotion Mixing Tips

Energy Combinations

• calm + confident = assured
• excited + happy = ecstatic
• tired + defeated = hopeless

Social Dynamics

• confident + mocking = smug
• nervous + apologetic = submissive
• angry + contemptuous = hateful

Physical + Emotional

• breathless + excited = exhilarated
• drunk + happy = jolly
• whispering + fearful = terrified

Character Voices

• Confident Scientist: confident + condescending + dismissive
• Nervous Assistant: nervous + insecure + apologetic
• Anxious Student: anxious + overwhelmed + stressed

📊 Parameter Guide

Emotion Intensity

• 0.0x: Neutral (no emotion)
• 0.5x: Subtle hint
• 1.0x: Normal strength (default)
• 1.5x: Strong emotion
• 2.0x: Extreme/exaggerated

How It Works

The node adjusts three generation parameters based on emotion:

• Temperature: Controls randomness/variation
• Repetition Penalty: Affects word repetition and flow
• Top-p: Nucleus sampling probability

Emotions are mixed by averaging their parameter adjustments, then multiplied by intensity.

🔧 Troubleshooting

Audio cuts off abruptly

• This is seed-dependent behavior
• Try randomizing seed multiple times
• Use longer input text
• Consider using x_vector_only=False with transcript

Generation too fast (3 seconds)

• This is normal for x_vector_only=True mode
• For longer, more accurate generation:
  • Set x_vector_only=False
  • Fill in ref_audio_transcript
  • Should take 15-20 seconds

Voice doesn't match reference

• Make sure reference audio is clear (5-30 seconds)
• Try x_vector_only=False with accurate transcript
• Use higher quality reference audio (24kHz recommended)

Node doesn't appear

• Verify base ComfyUI-Qwen-TTS is installed
• Check __init__.py was updated correctly
• Look for errors in ComfyUI console
• Try restarting ComfyUI

📝 Examples & Workflows

Sarcastic Scientist

Reference Audio: scientist_voice.mp3
Text: "Oh wonderful, another one who thinks they understand quantum mechanics."
Primary: sarcastic
Secondary: condescending
Tertiary: dismissive
Intensity: 1.5x

Defeated Character

Reference Audio: character_voice.mp3
Text: "I tried my best, but I guess it wasn't enough."
Primary: defeated
Secondary: sad
Tertiary: whispering
Intensity: 1.8x

Panicked Character

Reference Audio: anxious_voice.mp3
Text: "We need to move NOW, there's no time!"
Primary: panicked
Secondary: breathless
Tertiary: stressed
Intensity: 2.0x

🤝 Contributing

Contributions are welcome! Areas for improvement:

• Additional emotion presets
• Better parameter tuning for specific emotions
• Character-specific preset packs
• Multi-language emotion mapping

📜 Credits

• Emotional Voice Clone Node: Developed with assistance from Claude (Anthropic)
• Base Qwen3-TTS: Alibaba Qwen Team
• ComfyUI Integration: flybirdxx
• ComfyUI: comfyanonymous

📄 License

GPL-3.0 License (matching base Qwen3-TTS package)

🔗 Links

• Qwen3-TTS Official Repo
• ComfyUI-Qwen-TTS Base Package
• ComfyUI

💬 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Discord: ComfyUI Community Server

---

Made with 🎭 for the ComfyUI community