SicariusSicariiStuff/Impish_Bloodmoon_12B

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:Sep 1, 2025License:apache-2.0Architecture:Transformer0.1K Open Weights Warm

Impish_Bloodmoon_12B by SicariusSicariiStuff is a 12B parameter generalist language model specifically optimized for creative tasks such as story writing, adventure, and roleplay. It features an "impish" personality, robust GEC, a powerful synonym engine, and the unique ability to one-shot roleplay datasets in JSON form. The model also includes multi-lingual support for Russian, Hebrew, and Japanese, and is trained on a monstrous dataset to enhance multi-character interactions, theory of mind, multi-agent game theory, and strategic deception.

Loading preview...

Impish_Bloodmoon_12B: A Creative Powerhouse

Impish_Bloodmoon_12B, developed by SicariusSicariiStuff, is a 12B parameter model designed with a distinct "impish" personality, making it particularly adept at creative writing, adventure generation, and roleplay scenarios. It stands out for its ability to handle complex narrative structures and character interactions, offering a less positivity-biased output compared to many other models.

Key Capabilities

  • Creative Task Specialization: Excels in story writing, adventure generation, and roleplay, with a "wicked" innate personality.
  • One-Shot JSON Dataset Creation: Uniquely capable of generating JSON-formatted roleplay datasets in a single shot, beneficial for beginner model tuners.
  • Advanced Language Features: Includes robust Grammatical Error Correction (GEC) and a powerful synonym engine.
  • Multilingual Support: Supports Japanese, Hebrew, and Russian, with extensive training on these languages.
  • Enhanced Narrative Understanding: Features vastly improved multi-character interactions, theory of mind, multi-agent game theory, strategic deception, and item/stat tracking over long contexts.
  • Reduced Slop & High Diversity: Offers reduced "slop" in creative outputs and high swipe diversity for varied responses.

Good for

  • Developers and writers seeking a model with a distinct, less censored personality for creative projects.
  • Generating complex roleplay scenarios and adventure narratives.
  • Creating JSON-formatted datasets for model tuning, especially for those new to the field.
  • Applications requiring multi-lingual creative text generation in Japanese, Hebrew, or Russian.
  • Use cases demanding strong theory of mind and strategic interaction capabilities.

Finetunes
14 models