MN-VelvetCafe-RP-12B: A Merged Model for Immersive Roleplay
MN-VelvetCafe-RP-12B is a 12 billion parameter language model created by IggyLux through a SLERP merge of two specialized models: PocketDoc/Dans-PersonalityEngine-V1.3.0-12b and kyx0r/Neona-12B. This merge was specifically designed to enhance multi-character roleplaying experiences, focusing on visual detail and consistent narrative flow, particularly for users with VRAM limitations.
Key Capabilities
- Strong Visual Consistency: Excels at tracking and detailing scene elements, character positions, and clothing across multiple turns, preventing common immersion breaks.
- Balanced Emotional Responses: Provides nuanced emotional reactions without sudden, unprompted aggression or refusal, adapting to the narrative context.
- Reliable Instruction Following: Effectively processes author's notes and system prompts, ensuring the model adheres to user-defined guidelines.
- Adaptive Writing Style: Combines Neona's adaptability to writing styles and instruction following with Dan's focus on descriptive roleplay.
Good for
- Multi-Turn Group Roleplay: Optimized for scenarios involving two or more characters, maintaining coherence and detail.
- Visually Descriptive Narratives: Ideal for users who prioritize rich, detailed descriptions of environments, actions, and character appearances.
- Customizable Roleplay: Designed to be adaptable for both SFW and NSFW roleplay, maintaining emotional stability and instruction adherence.
- Users with 8GB VRAM: Developed with resource efficiency in mind, making it suitable for local deployment on systems with limited VRAM.