PersonalityPlex builds upon PersonaPlex to allow you to create (and share) Personalities that are used in PersonaPlex. PersonalityPlex provides the following features:
- Voice cloning from ~10s audio clips
- Create a library of Personalities
- Talk to your Personalities
- Make edits to you Personalities for better behavior
- Clean UI
Main Screen |
Personality Screen |
Clone a Voice |
Talk to your Personality |
See the Full Installation Guide (venv) to get up and running.
- Open your browser and go to https://localhost:8998
- Click Settings at the bottom of the left sidebar
- Enter the paths for the three required models:
- Moshi Weights - path to the language model checkpoint
- Mimi Weights - path to the audio codec model
- Text Encoder - path to the tokenizer model
- Click Save, then wait for loading to complete (the status indicator turns green when ready). This step will be skipped if the paths have already been set.
- In the left sidebar, click the dropdown menu and select Create a voice embedding
- Click Select audio file and choose a voice sample (~10 seconds or more)
- Enter a name for the embedding and click Generate Embedding
- Optionally, type some test text and click Test Embedding to hear a preview of the cloned voice
- The embedding will be stored in the
./Personalities/Embeddingsfolder
- In the left sidebar, click the dropdown menu and select Create new personality
- Fill in the personality details:
- Avatar - click to upload a custom image
- Name - give your personality a name
- Role (System Prompt) - describe how the personality should behave (e.g. "You are a pirate captain who loves telling sea stories")
- Embedding - select the voice embedding to use
- Adjust generation parameters if desired (text/audio temperature, topK, seed)
- Click Save
- The Personality will be stored in the
./Personalitiesfolder (an exmple Personality for Pam Beasley can be found on the Release Page)
- Click a personality in the left sidebar to select it
- Click Talk to your Personality
- Grant microphone permission when your browser prompts you
- Start speaking — the personality responds in real-time with audio and text
- The conversation is full-duplex, meaning you can speak and listen at the same time
- Click Disconnect when you're done



