Retreatcost/KansenSakura-Conflagration-RP-12b
KansenSakura-Conflagration-RP-12b by Retreatcost is a 12 billion parameter model, a successor to the Erosion engine, specifically optimized for kinetic narrative exchange in roleplay scenarios. This model prioritizes rapid-fire dialogue, decisive character actions, and intense interactive scenes over psychological introspection. It is designed for high-tension action, romantic or adversarial chemistry, and ensemble casts with distinct character voices, including NSFW capabilities.
Loading preview...
KansenSakura-Conflagration-RP-12b: Kinetic Narrative Engine
This model, developed by Retreatcost, is a 12 billion parameter language model designed as the "white-hot successor" to the Erosion engine. It trades deep introspection for "incineration" and "pyrotechnics," focusing on raw interactive intensity. The model is the first trained by its creator, indicating it may have unique quirks.
Key Capabilities & Characteristics
- Optimized for Kinetic Narrative: Specializes in rapid-fire dialogue and decisive character actions.
- High-Intensity Interactions: Excels in high-tension action sequences and dynamic character chemistry (romantic or adversarial).
- Character-Driven: Designed for ensemble casts with clear, reactive character voices.
- Reduced Introspection: Prioritizes immediate action and dialogue over psychological depth.
- NSFW Capabilities: Explicitly notes support for passionate/aggressive interactions.
- Advanced Merging Techniques: Built using complex
arcee_fusion,ties,karcher,multislerp,delerp, andnearswapmerge methods, including "reverse-abliteration" to enhance uncensored and dark outputs.
Optimal Use Cases
- Rapid-fire character banter and dialogue.
- High-tension action sequences.
- Romantic or adversarial chemistry scenarios.
- Scenes requiring decisive character action.
- Ensemble casts with clear character voices.
Users should note its heightened character reactivity and potential for scene momentum override. Recommended inference settings include a temperature of 0.8, repetition penalty of 1.05, and ChatML template format.