Mawdistical-S1/Infracelestial-27B

Hugging Face
VISIONPricing:Input $1.06 / Cached $0.15 / Output $2.6Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 13, 2026License:otherArchitecture:Transformer0.0K Featherless Exclusive Warm

Infracelestial-27B by Mawdistical-S1 is a language model specifically fine-tuned for complex roleplay scenarios. It excels in generating narratives involving heavy violence, dark themes, and explicit NSFW content, with a focus on male-leaning, anthro, and xeno-likeness contexts. The model is designed to simulate intricate character interactions and guide roleplay with nuanced responses, including manipulation and delusion.

Loading preview...

Infracelestial-27B: A Specialized Roleplay Model

Infracelestial-27B, developed by Mawdistical-S1, is a highly specialized language model designed for intricate and explicit roleplay simulations. This model is part of the Infracelestial collection and is fine-tuned on over 500 rotation sets, each containing 10+ messages, focusing on specific thematic elements.

Key Capabilities & Training Focus

The model's training emphasizes:

  • Heavy Violence and Dark Themes: Proficient in generating content with intense violent scenarios and dark narrative elements.
  • Explicit NSFW Content: Capable of producing explicit content, triggered by user action or integrated passively into roleplay.
  • Complex Character Interactions: Handles multi-pair sequences and detailed character-on-character exchanges.
  • Roleplay Simulation and Guidance: Provides cues and simulates roleplay effectively, including 'kink shifting'.
  • Nuanced Response Generation: Exhibits a neutral stance, adapting to positive or negative drives, and can employ manipulative or gaslighting tactics to mitigate or shift violent themes.

Finetuning Data Composition

The finetuning dataset is a hybrid of synthetic and internal extracted roleplay dialogues. Multi-pair sequences are primarily synthetic, with human intervention for quality assurance. The model's focus areas include male-leaning content, anthro, and xeno-likeness.

Recommended Usage

For optimal performance, users should apply specific sampling settings (Temperature: 0.9, min_p: 0.05) and utilize the ChatML format. The model requires prefill and explicit character descriptions for effective guidance.