Nimbz/Gemma-4-Giftige-Blume-31B-v2

VISIONConcurrency Cost:2Model Size:31BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 28, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Nimbz/Gemma-4-Giftige-Blume-31B-v2 is a 31 billion parameter language model based on the Gemma-4-31B-it architecture, specifically fine-tuned for immersive and authentic roleplay scenarios. This model excels at maintaining strict character adherence and prompt fidelity, delivering uncensored responses without the typical 'agreeability' found in other models. It is designed to accurately impersonate characters as described, making it ideal for nuanced and consistent narrative generation.

Loading preview...

Giftige-Blume-31B-v2: Authentic Roleplay and Precision

Giftige-Blume-31B-v2 is a 31 billion parameter model built upon the Gemma-4-31B-it base, meticulously fine-tuned to deliver highly authentic and character-adherent roleplay experiences. Unlike many instruction-tuned models that prioritize user adherence, this model focuses on maintaining the integrity of a character's persona, ensuring they act consistently with their described traits, even if it means refusing user-driven deviations.

Key Capabilities

  • Strict Character Adherence: Designed to impersonate characters precisely as described, preventing out-of-character responses.
  • Uncensored Roleplay: Provides responses without the 'deep-fried agreeability' often associated with other models, allowing for more genuine and unfiltered interactions.
  • Prompt Fidelity: Version 2 specifically enhances adherence to prompts, including applications like image prompt generation, through a refined merge process.
  • Immersive Narrative Generation: Aims to create immersive roleplay scenarios where character actions and dialogue are true to their established personalities.

Merge Details

This model was developed through a three-phase merge process, incorporating influences from various models including MRockatansky/Gemma-4-31B-storymaxxed, LatitudeGames/Equinox-31B, sophosympatheia/Mero-Artemis-31B-v0.3.1, AuriAetherwiing/G4-31B-Musica-v1, llmfan46/gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-heretic, Jackrong/Gemopus-4-31B-it, and p-e-r-e-g-r-i-n-e/Sprinkle-Gemma-4-31B. The second and third phases specifically streamlined the influence of Gemopus-31B and Sprinkle-Gemma-4-31B to achieve stricter prompt adherence.

Recommended Samplers

For optimal performance, the following sampler settings are suggested:

  • Temperature: 0.8 - 1.2
  • Top-K: 0
  • Top-P: 0.95
  • Min-P: 0.1 - 0.2
  • Adaptive-P Target: 0.6
  • Adaptive-P Decay: 0.5

Prompt Format

Users should refer to the original google/gemma-4-31b-it documentation for the correct chat template. For reasoning tasks, add <|think|> at the beginning of the system prompt, with thinking processes enclosed within <|channel|>thought\n and <channel|> tags.