whoashish115/Kitsune-Tales-E4B-EN

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 30, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Kitsune-Tales-E4B-EN by whoashish115 is an effective-4B parameter model (Gemma 4 E4B base) fine-tuned for generating original, general-audience fantasy light-novel fiction in English with Japanese anime themes. This 7.9 billion parameter model excels at creating synopses, short stories, and continuations from genre tags and a title, leveraging a LoRA SFT + DPO fine-tuning approach. It is specifically designed for creative writing and brainstorming within the fantasy light-novel genre, offering a 32768 token context length.

Loading preview...

Kitsune-Tales-E4B-EN: English Fantasy Light-Novel Generator

Kitsune-Tales-E4B-EN is an effective-4B parameter model (based on google/gemma-4-E4B-it) developed by whoashish115, specifically fine-tuned for generating original, general-audience fantasy light-novel fiction in English. This model, with 7.9 billion parameters and a 32768 token context length, specializes in producing content with Japanese anime and light-novel themes.

Key Capabilities

  • Original Fantasy Fiction: Generates synopses, short stories, and continuations based on genre tags and a title.
  • Japanese Anime/Light-Novel Themes: Incorporates stylistic elements and tropes common in Japanese anime and light novels.
  • Safety & Content Control: Trained to avoid sexual content, real people, and existing copyrighted characters, transposing non-fantasy requests into fantasy.
  • Improved Adherence: Demonstrates significantly higher length and tag adherence compared to its base model, with reduced degenerate outputs and markdown artifacts.

Training and Differentiation

This model is a LoRA SFT + DPO fine-tune of google/gemma-4-E4B-it, trained on filtered, cross-labeled synthetic data from two Apache-2.0 models. It is the English counterpart to Kitsune-Tales-E4B-JP, utilizing the same training methodology. The DPO phase notably improved quality and safety metrics. Automatic metrics show strong performance in genre adherence, title reflection, and fantasy-only output, with a low rate of repetition.

Good for

  • Creative Writing: Generating original fantasy light-novel content for hobbyists and writers.
  • Brainstorming: Assisting with plot ideas, character development, and story outlines within the specified genre.
  • Research: Studying small-model creative-writing fine-tuning using synthetic data.

Finetunes
3 models