aifeifei798/DarkIdol-Qwen3.8-27B-v1.1

Hugging Face
VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 6, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

DarkIdol-Qwen3.8-27B-v1.1 by aifeifei798 is a 27 billion parameter Qwen3.8-based model specifically fine-tuned for high-fidelity cinematic roleplay, interactive creative writing, and complex narrative simulations. It features strict anti-puppeteering, organic character consistency, and literary vernacular across English and Chinese. This model is designed for deep immersion in fictional scenarios, supporting multimodal vision inputs and integrated chain-of-thought reasoning for nuanced narrative progression.

Loading preview...

DarkIdol-Qwen3.8-27B-v1.1: Specialized Roleplay & Narrative Engine

DarkIdol-Qwen3.8-27B-v1.1 is a 27 billion parameter model developed by aifeifei798, exclusively fine-tuned for high-immersion roleplay (TRPG, SillyTavern), interactive creative writing, and complex narrative simulations. Unlike general-purpose instruction models, DarkIdol prioritizes a "Roleplay-First" philosophy, aiming to deliver deep, consistent, and engaging fictional experiences.

Key Capabilities & Design Philosophy

  • Strict Anti-Puppeteering: Ensures the model acts solely as the non-player character (NPC), never dictating user actions or dialogue.
  • Organic Character Consistency: Maintains character motives, flaws, and allegiances, preventing generic or out-of-character responses.
  • Literary Vernacular & Cadence: Excels in genre-authentic prose for both English (e.g., Southern Gothic, Tech-Noir) and Chinese (e.g., classical court intrigue).
  • Integrated Cognitive Reasoning: Utilizes internal <think> tags for chain-of-thought processing, allowing for psychological analysis, inventory management, and narrative logic.
  • Multimodal Vision Native: Supports simultaneous image inputs via a paired mmproj vision tower for dynamic scene art and tabletop map analysis.
  • High-Speed Inference: Optimized for deployment with llama.cpp using Multi-Token Prediction (MTP) and Quantized KV Cache, achieving 80-100+ tokens/second with 64K context.

When to Use This Model

  • Interactive Fiction & Roleplay: Ideal for creating immersive text-based adventures, TRPGs, and character-driven narratives where consistent NPC behavior and rich prose are paramount.
  • Creative Writing Assistance: For authors seeking a collaborative engine that can generate genre-specific literary content and complex plot developments.
  • Simulating Fictional Worlds: When the primary goal is to explore and interact within a purely fictional construct, without reliance on real-world factual accuracy.

Important Considerations

This model is explicitly designed for fictional content only. It carries a critical disclaimer that all outputs are synthetic literary hallucinations and should never be relied upon for real-world guidance, facts, safety advice, or instruction. Its safety filtering is significantly reduced to facilitate creative freedom, meaning users must exercise caution and assume full responsibility for generated content.