CharGen v2: Step-by-Step Character Generation
CharGen v2, developed by kubernetes-bad, is a 7 billion parameter model specifically engineered to assist in creating detailed role-playing characters. It distinguishes itself from other models by generating character descriptions field-by-field in a conversational style, rather than producing the entire character at once. This approach, a significant improvement over v1, minimizes repetition and enables users to re-roll specific sections like 'Description', 'Scenario', 'Personality', 'First message', and 'Dialogue examples' based on the Tavern Character Card V2 specification.
Key Capabilities
- Dialogue-driven generation: Creates character elements one at a time, allowing for iterative refinement.
- Reduced repetition: The step-by-step process helps mitigate common LLM repetition issues.
- Partial re-rolls: Users can regenerate individual character fields without affecting others.
- Custom dataset training: Trained on a unique, manually curated dataset of over 16,000 character cards, including some NSFW content.
- Advanced grammar correction: Utilizes a pipeline combining Coedit and Llama2-based models for high-efficacy grammar correction.
- "De-slopped" base model: Built upon "Good Robot," a base model specifically trained to reduce "GPT-isms" and improve instruction following.
Good For
- Role-playing enthusiasts: Ideal for users who need assistance in crafting rich and consistent character profiles.
- Creative writers: Helps overcome writer's block by providing a structured way to build character backstories and personalities.
- Interactive character editors: Its conversational format makes it suitable for integration into AI-powered character creation tools.