ReadyArt/Dark-Scarlett-v1.0-26B-A4B

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:2Model Size:26BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 14, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

ReadyArt/Dark-Scarlett-v1.0-26B-A4B is a 26 billion parameter Gemma-4 based language model fine-tuned by ReadyArt with a 32768 token context length. It is specifically optimized for uncensored, adult-oriented roleplay and ERP (erotic roleplay) scenarios, focusing on a Male User to Female AI perspective. The model utilizes a distinct dataset inspired by Melody1437's spirit, enhanced with specialized scenario prompts and advanced character/emotional engines for immersive narrative experiences. It is designed for personal, non-commercial use and generates mature content.

Loading preview...

Dark Scarlett 26B A4B: Uncensored Adult Roleplay Model

ReadyArt/Dark-Scarlett-v1.0-26B-A4B is a 26 billion parameter model built on the Gemma-4 architecture, fine-tuned by ReadyArt for uncensored, adult-oriented roleplay (ERP). It features a 32768 token context length, enabling longer and more coherent narrative experiences.

Key Capabilities & Features

  • Uncensored & Adult Content: Explicitly designed for mature themes and interactions, with 100% uncensored output.
  • Immersive Roleplay: Utilizes proprietary Character Engine and Emotional Engine for consistent personality, realistic reactions, and dynamic emotional states.
  • Perspective Optimization: Specifically trained for Male User → Female AI interactions, ensuring consistent roleplay dynamics.
  • Enhanced Narrative Flow: Improved context retention and longer conversation threads, with automated detection and rewriting of repetitive phrases.
  • Group Chat Support: Includes training data for dynamic multi-user group chat scenarios.
  • Clean Output: Reasoning tags, em dashes, asterisks, and markdown formatting are stripped from the dataset for cleaner roleplay generation.

Training Details

The model was fine-tuned using LoRA (Low-Rank Adaptation) over 2 epochs with a LoRA rank of 32, targeting text layers only. The dataset comprises 12,211 specialized adult-oriented roleplay prompts, distinct from Melody1437 but inspired by its spirit.

Recommended Usage

  • Sampler Settings: Recommended parameters include Top-P: 0.92, Temperature: 1.0, Frequency Penalty: 0, Presence Penalty: 0.
  • SillyTavern Integration: Supports chat_template_kwargs: {"enable_thinking": true, "reasoning_effort": "medium"} for enhanced AI reasoning.
  • GGUF Quantizations: Available in Q4_K_M, Q5_K_M (recommended), Q6_K, and Q8_0 formats for local inference.

Important Considerations

  • Content Advisory: This model generates mature content and is intended for adults (18+) only.
  • License: Built upon the Apache 2.0 license, for personal, non-commercial use only.