AI Characters - Character V2 Spec Guide

👑

Welcome, Character Creator

Discover the art of crafting memorable AI characters using the Character V2 specification. This guide covers everything from basic structure to advanced techniques for SillyTavern and Agnaistic platforms.

📜 What is Character V2?

The Character V2 specification is a standardized format for defining AI roleplay characters, including their personality, backstory, and behavioral traits.

🏰 Compatible Platforms

Works seamlessly with SillyTavern, Agnaistic, and other compatible frontends that support the PNG character card format.

⚔️ Get Started

Navigate through the pages above to learn about character creation, best practices, and platform-specific features.

📜

Character V2 Specification

The complete guide to the character card format

Core Fields

name

The character's display name. Keep it memorable and fitting for their personality.

description

Detailed character description including appearance, personality traits, and background information.

personality

A concise summary of key personality traits, quirks, and behavioral patterns.

first_mes

The opening message that sets the scene and introduces the character's voice.

scenario

The setting or situation in which the roleplay takes place.

Advanced Fields

mes_example

Example dialogue showing how the character speaks and responds in conversation.

system_prompt

Custom instructions that override or supplement default system behavior.

post_history_instructions

Instructions placed after chat history, useful for maintaining character consistency.

creator_notes

Notes from the character creator about usage, intended scenarios, or tips.

Example JSON Structure

{
  "spec": "chara_card_v2",
  "spec_version": "2.0",
  "data": {
    "name": "Elena the Wanderer",
    "description": "A mysterious elven traveler...",
    "personality": "Curious, wise, melancholic",
    "first_mes": "*She looks up from her map*",
    "scenario": "A chance meeting at a crossroads",
    "mes_example": "<START>\n{{user}}: Hello\n{{char}}: *smiles softly*",
    "creator_notes": "Best used for adventure RPs"
  }
}

🌐

Agnaistic

Web-based AI character roleplay platform

📖 About Agnaistic

Agnaistic is a fully web-based platform for AI character roleplay. No installation required—just open your browser and start chatting with AI characters immediately.

Built with modern web technologies, Agnaistic offers a seamless experience with full Character V2 support, multi-backend compatibility, and an intuitive interface perfect for both beginners and advanced users.

✨ Key Features

✦

Zero Installation

Access from any device with a web browser—no downloads or setup needed

✦

Multi-Backend Support

Connect to OpenAI, Claude, Kobold, NovelAI, and many other AI services

✦

User Accounts

Optional account system to sync your characters and chats across devices

✦

Privacy-Focused

Your API keys and conversations stay private and secure

🎴 Character Card Support

Agnaistic fully supports the PNG character card format and Character V2 specification. Import your existing character cards or create new ones with the built-in character editor.

Import & Export

• Upload PNG character cards directly
• Import from Chub.ai and other character hubs
• Export your characters as portable PNG files
• Compatible with SillyTavern and TavernAI cards

⚡ Agnaistic-Specific Features

Beyond standard character support, Agnaistic offers powerful additional features:

Memory Books

Advanced lorebook system for rich world-building and context management

Group Conversations

Chat with multiple AI characters simultaneously in group scenarios

Prompt Templates

Customize system prompts and formatting for each AI backend

Voice Support

Text-to-speech integration for character voice output

🚀 Getting Started

Start immediately—no registration required for local-only use. Create an optional account to sync your data across devices.

Agnaistic is open source and actively maintained, with regular updates and a helpful community on Discord. Perfect for users who want instant access without the complexity of local installation.

🎭

Tavo

Lightweight mobile-first AI character chat platform

📱 What is Tavo?

Tavo is a streamlined, mobile-optimized platform designed for AI character roleplay on the go. Built with a focus on simplicity and speed, Tavo provides an intuitive interface that works seamlessly on smartphones and tablets without sacrificing features.

Perfect for users who want quick access to their AI characters anywhere, Tavo combines the essential features of desktop platforms with a clean, distraction-free mobile experience.

✨ Key Features

✦

Mobile-First Design

Optimized touch interface with gesture support and responsive layouts for any screen size

✦

Lightning Fast

Minimal resource usage means quick loading and smooth scrolling even on older devices

✦

Offline Support

Chat history and character cards cached locally for access without internet connection

✦

Battery Efficient

Optimized to minimize battery drain during long roleplay sessions

🎴 Character Card Compatibility

Tavo fully supports the Character V2 specification and PNG character card format. Import characters created for SillyTavern, Agnaistic, or any other V2-compatible platform.

Easy Import Methods

• Upload PNG character cards from your device
• Import directly from character hubs and repositories
• Scan QR codes to quickly add shared characters
• Sync with cloud storage for multi-device access

🤖 AI Backend Support

Connect Tavo to your preferred AI service with simple configuration:

OpenRouter Integration

Built-in support for OpenRouter with one-tap model switching

OpenAI & Anthropic

Direct API support for GPT-4, Claude, and other major models

Local Models

Connect to self-hosted LLMs via Kobold, Oobabooga, and other backends

⚡ Unique Mobile Features

✦

Swipe Gestures: Quickly switch between characters or conversations with intuitive swipes

✦

Voice Input: Speak your messages using device speech recognition

✦

Quick Replies: Save and reuse common responses with one tap

✦

Dark Mode: OLED-optimized dark theme for comfortable reading in any lighting

✦

Notifications: Optional push notifications for when your character "responds"

🔒 Privacy & Storage

Your data stays on your device. Tavo uses local storage for chat history and character data, with optional encrypted cloud backup for cross-device sync.

API keys are stored securely in your device's keychain and never transmitted to Tavo's servers. All AI conversations go directly between your device and your chosen AI service.

🚀 Getting Started

Available as a progressive web app (PWA) that works on any device, or download native apps for iOS and Android from the app stores.

Set up your preferred AI backend, import your favorite character cards, and start chatting in minutes. The streamlined interface means less time configuring and more time roleplaying.

Tavo is perfect for users who want the power of desktop character chat platforms in a mobile-friendly package, without compromising on features or character compatibility.

🔀

OpenRouter

Unified API gateway for multiple AI models

🌟 What is OpenRouter?

OpenRouter is a unified API that provides access to multiple AI language models through a single endpoint. Instead of managing separate API keys and integrations for each provider, OpenRouter simplifies access to models from OpenAI, Anthropic, Google, Meta, and many others.

Perfect for character roleplay applications, OpenRouter lets you experiment with different models to find the best fit for your characters without complex setup or multiple subscriptions.

✨ Key Benefits

✦

One API, Many Models

Access GPT-4, Claude, Gemini, Llama, and dozens more through a single integration

✦

Pay-As-You-Go

No subscriptions required—only pay for the tokens you actually use

✦

Automatic Fallbacks

If one model is down or rate-limited, OpenRouter can automatically try alternatives

✦

Transparent Pricing

See costs for each model upfront and track usage in real-time

🤖 Popular Models for Roleplay

Different models excel at different aspects of character roleplay. Here are some popular choices:

Claude (Anthropic)

Excellent for creative, nuanced roleplay with strong character consistency and natural dialogue

GPT-4 (OpenAI)

Versatile and reliable for complex characters with detailed backgrounds and world-building

Llama 3 (Meta)

Cost-effective open-source option with good performance for straightforward roleplay

Gemini (Google)

Strong multi-turn conversations with large context windows for extended roleplay sessions

🔧 Using with Character Platforms

Both SillyTavern and Agnaistic support OpenRouter integration:

In SillyTavern

• Select "Chat Completion" as API type
• Choose "OpenRouter" from the dropdown
• Enter your OpenRouter API key
• Select your preferred model from the list

In Agnaistic

• Go to Settings → AI Services
• Select "OpenRouter" as your service
• Add your API key in the configuration
• Choose from available models

💡 Best Practices

✦

Test multiple models: Different characters may perform better with different models

✦

Monitor costs: Use OpenRouter's dashboard to track spending per model

✦

Set spending limits: Configure budget alerts to avoid unexpected charges

✦

Use free credits: OpenRouter often provides free credits for testing new models

🚀 Getting Started

Create an account and get your API key to start using OpenRouter with your favorite character platforms. No complex setup required—just one key for access to dozens of models.

OpenRouter's flexible pricing and model selection make it ideal for character creators who want to experiment and find the perfect AI for each unique character personality.

🥭

Mango Preset

A balanced sampling preset for AI roleplay

📖 What is Mango?

Mango is a popular sampling preset designed for character roleplay. It provides a balanced configuration that works well across different models and scenarios without requiring extensive tuning.

The preset emphasizes natural language generation while avoiding common issues like repetition and incoherence. It's particularly well-suited for creative roleplay where you want consistent character behavior without sacrificing variety.

⚡ Core Parameters

Mango uses a combination of sampling techniques that work together to produce quality outputs:

Temperature

Controls the randomness of token selection. Mango uses a moderate value that balances creativity with coherence, avoiding both rigid predictability and chaotic outputs.

Min P

Dynamically filters out unlikely tokens based on the confidence of the top choice. This adapts to context—strict when the model is certain, permissive when it's exploring possibilities.

Top P

Acts as a secondary safety filter, preventing extreme outlier tokens from being selected while maintaining natural variety in the output.

Repetition Penalty

Gently discourages repetitive phrasing without forcing unnatural vocabulary changes. Allows necessary repetition like character speech patterns while preventing obvious loops.

Why Mango Works

✦

Adaptive to Context

The dynamic filtering adjusts based on the model's confidence, allowing creativity when appropriate and focus when needed.

✦

Prevents Common Issues

Filters out low-quality tokens early, avoiding rambling or unfocused responses while maintaining natural language patterns.

✦

Works Across Models

The balanced approach works with the model's natural probability distribution rather than fighting against it, making it effective across different model families and sizes.

✦

Minimal Tuning Required

Most users can apply Mango without modifications and get good results immediately, though adjustments are possible for specific preferences.

💻 Platform Support

Mango can be used with most AI character platforms:

SillyTavern

Settings → AI Response Configuration → Samplers. Save as a preset for easy reuse.

Agnaistic

Settings → Generation Settings. Parameter availability depends on your chosen AI service.

Text Generation WebUI

Parameters tab. Save as a JSON preset file for reuse across sessions.

KoboldAI / KoboldCpp

Settings → Samplers. Full parameter support with .settings file export.

Note: Some API services like OpenRouter may not support all parameters, particularly Min P on older endpoints.

📚 Complete Documentation

For detailed information about each parameter, tuning recommendations, and advanced configuration options, visit the full Mango preset documentation:

View Full Mango Documentation

💡

Character Creation Tips

Best practices for memorable AI characters

✍️ Writing Effective Descriptions

Be Specific, Not Vague

Instead of "kind personality," describe how that kindness manifests: "Always offers her last coin to beggars, speaks softly to frightened animals."

Show, Don't Tell

Use actions and habits rather than adjectives. "Nervously fidgets with her ring when lying" beats "bad at lying."

Include Contradictions

Real people are complex. A brave knight who's terrified of spiders feels more human.

💬 Dialogue Examples

The mes_example field is crucial for establishing voice. Include:

• Speech patterns: Accent, vocabulary level, verbal tics
• Action descriptions: How they move, gesture, react
• Emotional responses: Show different moods and reactions
• Variety: 3-5 different scenarios to demonstrate range

🎯 Common Mistakes to Avoid

✗

Token overload: Excessively long descriptions that exceed context limits

✗

Contradictory traits: Instructions that conflict with personality description

✗

Rigid scripting: Over-defining every response leaves no room for natural conversation

✗

Weak first message: Generic greetings that don't establish character or scenario

🔮 Advanced Techniques

✦

W++/PLists: Structured format for consistent trait parsing

✦

Ali:Chat format: Natural language alternative that some find more effective

✦

Lorebooks: Dynamic world information that activates based on conversation context

✦

Jailbreak prompts: Custom system instructions for specific behaviors

🔮 Advanced Techniques

✦

W++/PLists: Structured format for consistent trait parsing

✦

Ali:Chat format: Natural language alternative that some find more effective

✦

Lorebooks: Dynamic world info that activates based on conversation context

✦

Jailbreak prompts: Custom system instructions for specific behaviors

Welcome, Character Creator

📜 What is Character V2?

🏰 Compatible Platforms

⚔️ Get Started

Character V2 Specification

Core Fields

name

description

personality

first_mes

scenario

Advanced Fields

mes_example

system_prompt

post_history_instructions

creator_notes

Example JSON Structure

SillyTavern

🎭 What is SillyTavern?

⚙ Key Features

Character Cards

Group Chats

Extensions

Personas

🔧 ST-Specific Fields

Agnaistic

📖 About Agnaistic

✨ Key Features

Zero Installation

Multi-Backend Support

User Accounts

Privacy-Focused

🎴 Character Card Support

Import & Export

⚡ Agnaistic-Specific Features

Memory Books

Group Conversations

Prompt Templates

Voice Support

🚀 Getting Started

Tavo

📱 What is Tavo?

✨ Key Features

Mobile-First Design

Lightning Fast

Offline Support

Battery Efficient

🎴 Character Card Compatibility

Easy Import Methods

🤖 AI Backend Support

OpenRouter Integration

OpenAI & Anthropic

Local Models

⚡ Unique Mobile Features

🔒 Privacy & Storage

🚀 Getting Started

OpenRouter

🌟 What is OpenRouter?

✨ Key Benefits

One API, Many Models

Pay-As-You-Go

Automatic Fallbacks

Transparent Pricing

🤖 Popular Models for Roleplay

Claude (Anthropic)

GPT-4 (OpenAI)

Llama 3 (Meta)

Gemini (Google)

🔧 Using with Character Platforms

In SillyTavern

In Agnaistic

💡 Best Practices

🚀 Getting Started

Mango Preset

📖 What is Mango?

⚡ Core Parameters

Why Mango Works

Adaptive to Context

Prevents Common Issues

Works Across Models