DarwinAnim8or/Trouper-v2-12b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

DarwinAnim8or/Trouper-v2-12b is a 12 billion parameter role-playing (RP) model fine-tuned on Mistral Nemo Base, designed for character writing that produces clean, human-like prose. It excels at generating nuanced character responses with internal reasoning ('think blocks') and maintains user agency by not narrating user actions. This model is specifically optimized for rich, long-running character interactions and handles adversarial prompts robustly, making it ideal for immersive storytelling applications.

Loading preview...

Trouper-v2-12B: Advanced Role-Playing Model

Trouper-v2-12B is a 12 billion parameter model, fine-tuned on Mistral Nemo Base, specifically engineered for high-quality character role-playing. It represents a ground-up rebuild from its predecessor, focusing on generating prose that emulates human writing rather than typical AI output. The model's core philosophy emphasizes clean prose, strong character voice differentiation, and realistic pacing in interactions.

Key Capabilities & Features

  • Thinking Characters: Incorporates 'think blocks' that simulate a character's internal monologue, driving more intentional and grounded responses without leaking into visible prose.
  • Restraint & Pacing: Trained on rich character cards with explicit arc stages, allowing characters to develop warmth gradually and hold back appropriately, avoiding premature emotional reveals.
  • Adversarial Robustness: Handles challenging user inputs, absurd requests, and boundary pushes in-character, maintaining narrative consistency.
  • User Agency: Strictly adheres to character perspective, never narrating user actions, emotions, or decisions.
  • Rich Card Support: Utilizes structured character cards with facade, hidden depth, voice, and tells for nuanced character portrayal, while also gracefully handling simpler 'vibe cards'.
  • SillyTavern Utility Prompts: Natively supports out-of-band calls for story summaries and image-generation prompts within SillyTavern, eliminating the need for a secondary model.
  • Low Slop & Repetition: Benchmarked against slop-forensics, Trouper-v2-12B demonstrates significantly lower repetition rates compared to other models in its class, ensuring varied and engaging dialogue.

Ideal Use Cases

  • Immersive Role-Playing: Best suited for applications requiring deep, consistent character interactions and long-running narrative sessions.
  • Creative Writing & Storytelling: Excellent for generating character-driven narratives with natural prose and nuanced emotional development.
  • SillyTavern Integration: Optimized for seamless use within SillyTavern for character AI, including its unique utility prompt handling.
  • Single-GPU Setups: Offers fast inference speeds and lower VRAM requirements (~8GB quantized) compared to larger models like Prima-v2-24B, making it accessible for many users.