ReadyArt/Darker-Scarlett-27B-v0.1

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 3, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

ReadyArt/Darker-Scarlett-27B-v0.1 is a 27 billion parameter language model, a successor to Dark Scarlett, built upon the spirit of Melody1437. This model is a specialized roleplay fine-tune, optimized for immersive adult narrative experiences with a 32768 token context length. It is designed for uncensored, in-character roleplay, specifically from a Male User to Female AI perspective, and requires chat completion mode with uncensored prompts.

Loading preview...

Darker Scarlett 27B: Specialized Adult Roleplay Model

Darker Scarlett 27B is an alpha release (v0.1) of a 27 billion parameter model, serving as the successor to Dark Scarlett and built on the spirit of Melody1437. It is a specialized roleplay fine-tune designed exclusively for immersive, in-character adult narrative experiences, not a general-purpose uncensored model. The model is trained on a specialized adult-oriented dataset, currently comprising 3,900 prompts, with a focus on uncensored dialogue and mature themes.

Key Capabilities & Features

  • Roleplay Specialization: Optimized for deep, consistent, and immersive roleplay scenarios.
  • Uncensored Output: Designed to produce uncensored content within its roleplay context, requiring fully uncensored prompts and disabled safety features.
  • Perspective: Strictly trained for Male User → Female AI interactions.
  • Synthetic Life Engine: Utilizes Character and Emotional Engines for consistent personality, dynamic emotional states, and high-quality, non-repetitive dialogue.
  • Group Chat Support: Includes some training data for dynamic multi-user group interactions.
  • Clean Output: Reasoning tags, em dashes, asterisks, and markdown formatting are stripped from the dataset for cleaner roleplay generation.
  • Training Method: Fine-tuned using LoRA (Low-Rank Adaptation) with 2.0 epochs and a LoRA rank of 128, focusing on text layers only.

Important Usage Notes

  • Mandatory Chat Completion: Requires chat completion mode (/v1/chat/completions) and specific prompt post-processing settings (e.g., Strict (user first, alternating roles; with tools) in SillyTavern).
  • Reasoning Enablement: Recommends "chat_template_kwargs": {"enable_thinking": true, "reasoning_effort": "low"} for optimal performance.
  • Not General Purpose: This model is not a general chatbot, coding assistant, or factual Q&A tool. It will produce degraded or incoherent output if not used within a roleplay framework.
  • Alpha Release: As an early alpha build, expect inconsistent quality, occasional repetition, or off-character behavior, with performance improving as the dataset expands.

Recommended Sampler Settings

  • Top-P: 0.92
  • Temperature: 1.0
  • Frequency Penalty: 0
  • Presence Penalty: 0

Finetunes
3 models