How to Use Custom LLMs to Build Lifelike Virtual 3D Characters

By
Convai Team
January 7, 2025

Have you struggled to bring depth and control to your virtual characters or found that existing AI solutions lack the flexibility you need? Creating lifelike, interactive characters is challenging, especially when you want them to respond intelligently and responsibly in diverse situations. Balancing safety, customization, and character personality can be an ongoing challenge for virtual world creators, developers, and designers.

With Convai's new Core AI Settings, we’re giving creators the power to design characters that are not only smart but also adaptable and safe. These settings let you control key aspects like selecting custom language models to power characters, enabling moderation filters, and controlling response creativity. It provides an unprecedented level of customization to suit your unique project needs.

Until now, creators have struggled to balance customization with safety and realism in character interactions. Without the right tools, maintaining control over language, response tone, and content moderation can be challenging, leading to characters that don’t quite hit the mark or engage users fully. 

Core AI Settings addresses these challenges by offering a streamlined, intuitive way to manage every aspect of your character's interactions, ensuring they deliver memorable experiences while staying aligned with your goals. Let’s see how.

Convai’s Core AI Settings dashboard with Convai Playground.

Meet Convai's Core AI Settings: Customize Your Virtual Characters Like Never Before

Each Core AI Setting offers options that help creators fine-tune every interaction, set boundaries for safe conversations, and customize character responses to create immersive experiences that feel true to life

Here’s a closer look at what each setting does, why it’s groundbreaking, and how to use it to level up your virtual characters.

Keep Interactions Safe with the Moderation Filter

The Moderation Filter helps maintain a respectful, safe conversation environment by preventing inappropriate content from appearing in the chat, including hate speech, profanity, and other harmful language. This feature is essential for public-facing applications, educational platforms, or any environment where user safety is a priority.

Unlike traditional moderation tools, Convai’s Moderation Filter is tightly integrated with character functionalities like Narrative Design and Multilingual support

Turning on this filter ensures that your character complies with moderation standards while unlocking additional capabilities that rely on safe, structured language. It allows for worry-free public interactions without constant oversight.

How It Works: Simply toggle the Moderation Filter on or off in the Core AI Settings panel. Disabling it will make certain features—like Narrative Design and Multilingual capabilities—unavailable, as these rely on moderated, controlled language for smooth performance.

GIF showing how to toggle the Moderation Filter on and off in the Core AI Settings interface in Convai Playground.

➡️ Suggested Resource: Experience Safety Overview: Understanding Guardrails for Convai-Powered AI Characters.

Choose the Right Model for Every Interaction

Within Core AI Settings, you can choose from a diverse range of large language models (LLMs) to create a character that feels just right for your use case. Each model has distinct characteristics, allowing you to pick one that best matches your character’s desired tone and responsiveness. 

Here’s the comparison table with speed, style, cost, censorship status, and best use case for each foundation model in Convai’s Core AI Settings so you can make informed decisions when selecting foundation models:

Foundation Models Table

Foundation Models

Foundation Model Model Code Speed Style Cost (per 1M output tokens) in USD Best Use Case Uncensored
GPT-4o gpt-4o Fast Highly Conversational, Creative $10 Advanced conversational applications requiring creativity No
GPT-4o-mini gpt-4o-mini Very Fast Fast, Compact $0.60 Compact, quick responses for small tasks No
Claude 3.5 Sonnet claude-3-5-sonnet Moderate Balanced, Long-form Responses $15 Long-form answers or structured narratives No
Gemini-1.5-pro gemini-1.5-pro Moderate Fast, Efficient $2.19 General-purpose applications with faster responses No
Gemini-1.5-flash gemini-1.5-flash Very Fast Quick, Compact Responses $0.38 Quick replies in time-sensitive environments No
LLaMA3-70B llama3-70b Medium Balanced, Factual $0.89 Balanced factual information and conversational ability Yes
LLaMA2-13B llama-2-13b Slow Concise, Optimized $0.50 Lightweight use cases with basic responses Yes
Fine-Tuned Mistral-7B uncensored-small Fast Efficient, Tuned for Specific Tasks $0.25 Specific tasks requiring domain-specific fine-tuning Yes
Fine-Tuned GPT-3.5-Turbo N/A Very Fast Balanced, Highly Optimized Balanced tasks requiring high optimization and efficiency No

Each model brings a unique "voice" to interactions, so you can select one that best fits the character’s purpose and audience. This flexibility is especially valuable for applications ranging from customer support to interactive storytelling.

Check out more details in the documentation.

How It Works:

In the Core AI Settings panel, simply use the dropdown menu to select from models like GPT-4o, Claude 3.5 Sonnet, Gemini-1.5-pro, and more. If the Moderation Filter is turned off, certain models may become unavailable, which helps maintain content safety across interactions.

A GIF showcasing the range of model options, highlighting popular choices like GPT-4o and Gemini-1.5-pro.

Control Your Character’s Creativity with the Temperature Setting

The Temperature setting allows you to adjust the level of randomness and creativity in your character’s responses:

  • Higher temperatures yield more varied and creative replies, perfect for storytelling or exploratory conversations.
  • Lower temperatures produce more focused and consistent responses, which is ideal for structured applications like customer service or educational content.

Temperature adjustment is excellent for creators who want to fine-tune character responses without switching models. It gives them nuanced control over interaction style, allowing them to adjust character behavior to fit different contexts, from interactive fiction to customer service applications.

How It Works:

Adjust the Temperature slider to the desired level in the Core AI Settings panel. Higher settings (closer to 1.0) introduce more diversity in responses, while lower settings (near 0) produce reliable, predictable replies. Experiment with this setting to find the best balance for your use case.

A GIF showing a close-up of the Temperature slider, showing how it’s adjusted.

By bringing these customizable settings to Convai’s platform, we give creators the tools to craft unique, compelling, and safe virtual characters. 

Whether designing for a game, an educational platform, or a customer support role, Core AI Settings allows you to easily shape every aspect of your character’s personality and interaction style.

Why Core AI Settings Are a Game-Changer for Creators

Convai’s Core AI Settings bring several advantages to the development and creative process, making them an invaluable toolkit for building engaging, customizable, and safe virtual characters:

  1. Total Customization, Your Way: Core AI Settings allows you to control the specifics of character interactions in real-time. From moderation to model selection and response style, these settings empower you to bring your unique vision to life.
  2. Craft Immersive Experiences with Confidence: By fine-tuning each setting, you can ensure that characters provide an engaging, realistic, and safe experience. These settings help shape a more satisfying user experience for entertainment or education.
  3. Versatility Across Applications: The settings cater to a wide range of use cases—from games to customer support to virtual training environments—empowering creators to adapt characters to the needs of any scenario.

How to Select a Custom Large Language Models (LLMs) for Your Virtual AI Characters

Getting started with Core AI Settings is simple. Here’s how to begin exploring these features:

1. Explore in the Playground:

You can access Core AI Settings directly through Convai’s Playground, where you can test and tweak settings in a live environment. The Playground is designed to help you quickly understand how each feature impacts your character’s interactions.

2. API Access for Advanced Users

For those looking to integrate these features programmatically, Convai’s API gives you full access to Core AI Settings. This is ideal for larger projects or applications where real-time adjustments and scalability are crucial.

import requests
import json

url = "https://api.convai.com/character/update"

headers = { 
    'CONVAI-API-KEY': '<Your-API-Key>',
    'Content-Type': 'application/json'
}

# Create a dictionary for the JSON payload
payload = { 
    "charID": "<Your-Character-Id>",
    "model_group_name": "claude-3-5-sonnet",
    "temperature": 0.89
}

# Convert the payload to JSON
json_payload = json.dumps(payload)

response = requests.post(url, headers=headers, data=json_payload)

print(response.text)

Conclusion and Next Steps: How to Use Custom LLMs to Build Lifelike Virtual 3D Characters

Convai's Core AI Settings empower you with customization power, simplifying the process of creating virtual characters that align with your vision. 

With control over moderation, model selection, and creativity (temperature), you can craft experiences that are as safe, engaging, and unique as your users deserve. 

Your characters can thrive in any virtual environment by giving creators control over moderation, model selection, and temperature.

Ready to start building the next generation of virtual characters? Head to Convai's Playground today to explore Core AI Settings and take your virtual characters to the next level.