Nimbz/Gemma-4-Gembrain-X-31B
Nimbz/Gemma-4-Gembrain-X-31B is a 31 billion parameter merged language model built upon the Gemma 4 architecture, designed for advanced creative writing, nuanced roleplay, and precise image prompt generation. It features a dual-mode operation, offering fast general inference or deep, character-adherent reasoning, and maintains strong prompt adherence and swipe variety. The model excels at generating vivid descriptions and handling explicit content with descriptive precision, supporting a 32768 token context length.
Loading preview...
Overview
Nimbz/Gemma-4-Gembrain-X-31B is a sophisticated 31 billion parameter merge of the Gemma 4 architecture, meticulously rebuilt from the original Gembrain-31B. This model is engineered to offer a balance of creativity, intelligence, context-awareness, and prompt adherence, aiming to surpass its predecessor's capabilities, particularly in engaging and non-refusing interactions.
Key Capabilities
- Dual-Mode Operation: Operates in two distinct modes:
- Fast & Versatile (Thinking Off): Ideal for general questions, creative roleplay, casual conversations, and rapid, structured image prompt generation.
- Deep & Adherent (Thinking On): Activates enhanced reasoning for strict character adherence, complex tasks, and maintaining coherence in intricate scenarios. Reasoning is described as a blend of Gemini-style and Claude-style thinking.
- Creative Roleplay & Swipe Variety: Excels in open-ended roleplay, generating fresh scenes, scenarios, and varied character inventions without repetition. It maintains character agency and paces scenes effectively.
- Descriptive Precision: Capable of using diverse language registers and providing vivid descriptions, particularly noted for handling explicit content with detailed portrayal rather than generic euphemisms.
- Image Prompt Generation: Follows specified prompt structures to produce clean, vivid descriptions for characters or scenery, generating multiple distinct image prompts in a single response without losing coherence.
- Prompt Sensitivity: Adjusts output length and verbosity based on system prompt framing, adapting to conversational styles like roleplay or character-centered instructions.
- Long Context Consistency: Global-attention layers are shielded during merging to preserve reasoning and detail-tracking across long chats, tested up to 30k tokens.
Good For
- Developers seeking a highly steerable model for creative applications.
- Users requiring nuanced and varied outputs for roleplaying scenarios.
- Artists and designers needing precise and vivid image prompt generation.
- Anyone looking for a model that can adapt its output style and depth based on prompt instructions.