Skip to content

Mimic AI Digital Assistant v 1.2.1

Latest
Latest

Choose a tag to compare

View all tags
@bmerriott bmerriott released this 02 Mar 05:24
v1.2.1
0b8184a

icon

🎬 Demo Video

Watch the full demo of MIMIC AI v1.2.0 in action:

👉 MIMIC AI Desktop Assistant v1.2.0 - YouTube

Featuring voice cloning, multi-persona support, and real-time conversation

Mimic AI v1.2.0 - Voice Profiles & Enhanced TTS

🎉 Major Update

Mimic AI v1.2.0 introduces persistent voice profiles for instant voice cloning, improved TTS error handling, Qwen3 1.7B model support with VRAM management, and per-persona memory isolation.

📦 What's Included

**Updated to v1.2.1 to address crashes when accessing the memory manager.

File Description Size
Mimic-AI-Setup.exe One-click installer (recommended) ~164 MB
Mimic-AI-v1.2.0.zip Portable ZIP archive ~164 MB
launch-mimic.bat Launch with Python backend -
MimicAI-StopAll.bat Stop all services -
README.md Full documentation -
QUICKSTART.md Quick start guide -
CHANGELOG.md Version history -

🚀 Getting Started

New Users

  1. Download Mimic-AI-v1.2.0.zip from GitHub Releases
  2. Extract the entire ZIP to a folder
  3. Run Mimic-AI-Setup.exe
  4. Launch from Start Menu or Desktop
  5. Create your first persona with voice cloning!

Upgrading from v1.1.0?

Your existing personas, memories, and VRM avatars are preserved. Voice profiles will be created automatically when you first use voice features.

Prerequisites (Same as v1.1.0)

  1. Python 3.10-3.12

  2. Ollama (for AI models)

  3. Docker Desktop (optional, for web search)

✨ What's New in v1.2.0

🎙️ Voice Profile System (New!)

  • Persistent Voice Cloning - Create voice profiles that persist across sessions
  • Instant Voice Switching - Cached voice clone prompts for near-instant TTS
  • Voice Profile API - Save/load voice profiles per persona
  • Disk Caching - Voice clone prompts cached to disk for fast reloading

🧠 Per-Persona Memory Isolation (New!)

  • Independent Conversations - Each persona now has completely isolated chat history
  • Memory Files - Stored at ~/MimicAI/Memories/{persona_id}/
  • Auto-Clear on Switch - Chat automatically clears when switching personas
  • No Cross-Contamination - Persona A cannot see Persona B's conversations

🔊 Enhanced TTS Engine Support

  • Qwen3 1.7B Model - Higher quality voice generation (requires ~6GB VRAM)
  • Auto VRAM Detection - Automatically falls back to 0.6B if insufficient VRAM
  • GPU Memory Management - New /api/tts/unload endpoint to free VRAM
  • Engine Switching - Properly unloads models when switching between Qwen3 and KittenTTS

🛡️ Improved Error Handling

  • TTS Failure Resilience - Text response displays even when voice generation fails
  • Graceful Degradation - If Qwen3 fails, automatically shows text-only response
  • Better Error Messages - Clear feedback when tensor mismatches or VRAM issues occur
  • Voice Prompt Loading - Shows "Generating response..." indicator during LLM inference

🔧 Backend Improvements

  • Fixed Voice Clone Caching - Properly caches full VoiceClonePromptItem list
  • Reference Text Handling - Fixed reference text not being passed to voice creation
  • Memory Management - Added unload_qwen3() with CUDA cache clearing and garbage collection
  • VRAM Check - Pre-load VRAM check prevents OOM crashes

🎬 VRMA Library Cleanup

  • Removed Unused Animations - Cleaned up duplicate and unused VRMA files
  • Synced Resources - All VRMA files properly synced between app/public/vrma/ and src-tauri/resources/vrma/
  • Optimized Bundle - Reduced release package size

⚠️ Important: License

Mimic AI is licensed under a Proprietary License as a Freemium Product with In-app Purchases

📋 System Requirements

  • OS: Windows 10/11 (64-bit)
  • RAM: 8GB minimum, 16GB recommended
  • Storage: 5GB free space
  • GPU: NVIDIA GPU with CUDA support recommended for Qwen3 TTS
    • Qwen3 0.6B: ~3GB VRAM
    • Qwen3 1.7B: ~6GB VRAM

🐛 Known Issues

  • First launch may take 30-60 seconds as bundled assets are copied to user data
  • Docker must be running before enabling web search
  • Qwen3 1.7B requires significant VRAM - will auto-fallback to 0.6B if insufficient
  • Some antivirus software may flag the installer (false positive)

🔧 Troubleshooting

"Voice generation failed: Sizes of tensors must match"

  • This occurs with incompatible reference audio formats
  • Try re-recording your voice sample in Voice Creator
  • Ensure reference audio is clear and 3-10 seconds long

Qwen3 TTS not working / Out of Memory

  • Check GPU VRAM availability in Task Manager
  • The app will auto-fallback to 0.6B if 1.7B doesn't fit
  • Use POST /api/tts/unload to free VRAM when switching engines

Text not appearing with voice prompts

  • Fixed in v1.2.0 - voice prompts now show "Generating response..." indicator
  • Text appears synchronized with audio playback

All other issues

  • See v1.1.0 troubleshooting guide in README.md

🎨 Attribution

  • Default Avatar: PeePhanthong (VRoid Hub)
  • Animations: VRoid Project by pixiv Inc., 八ツ橋まろんのお店, Kannaku @ Nekokoya
  • TTS: Qwen3-TTS by Alibaba, KittenTTS by KittenML
  • See ATTRIBUTION.md for full credits

📄 License

Proprietary License - See LICENSE.txt for full terms

Full Documentation: See README.md and CHANGELOG.md
Issue Tracker: https://github.com/bmerriott/MIMIC-Multipurpose-Intelligent-Molecular-Information-Catalyst-/issues

Assets 4
Loading