High-fidelity character LoRA for Senko-san from The Helpful Fox Senko-san, for use with Krea 2 Turbo.Built from 1,000 native 1024 × 1024 NovelAI 5.0 Curated generations, this LoRA targets sharp anime images with defined linework, solid anatomy, and minimal compression blur. It is designed for high-quality <code>txt2img</code> character generation with strong identity consistency and flexible natural-language prompting.This SUPER version combines the 598-image base dataset used to make my NovelAI 4.5 Curated LoRAs with 402 new activity scenes, expanding scene and activity versatility while preserving Senko’s likeness.<h2 id="the-trigger-word">The Trigger Word</h2><pre><code>senko_san</code></pre>Include <code>senko_san</code> in your positive prompt, followed by a description of the outfit, activity, setting, and visual style you want.<h2 id="what-this-model-does-best">What This Model Does Best</h2><ul><li>Character identity: Reproduces Senko’s recognizable face, amber eyes, short layered golden-blonde hair, fox ears, fluffy fox tail, hair ornament, and anime appearance.</li><li>Outfit flexibility: Maintains her identity across dresses, casual clothing, jackets, hoodies, kimono, and other outfits.</li><li>Natural-language prompting: Supports descriptive prompts through Krea 2 Turbo’s Qwen3-VL-4B text encoder. Describe poses, expressions, clothing, lighting, and composition in clear sentences.</li><li>Scene flexibility: Handles indoor, outdoor, traditional, modern, daytime, sunset, and nighttime environments.</li><li>Fast generation: Uses Krea 2 Turbo’s 8-step generation workflow for quick iteration.</li></ul><h2 id="recommended-generation-settings">Recommended Generation Settings</h2>Use these settings as your out-of-the-box starting point in ComfyUI:Setting Recommendation Base model Krea 2 Turbo Text encoder Qwen3-VL-4B, loaded with type <code>krea2</code> VAE Qwen Image VAE Trigger word <code>senko_san</code> Prompt style Descriptive natural language LoRA weight Start at <code>1.0</code>; adjust within <code>0.85–1.0</code> Sampler Euler (<code>euler</code>) Scheduler Simple (<code>simple</code>) Steps <code>8</code> CFG scale <code>1.0</code> in ComfyUI Denoise <code>1.0</code> for <code>txt2img</code> Resolution Start at <code>1024 × 1024</code>; Turbo also supports generation up to <code>2048 × 2048</code> Negative prompt Leave empty at the recommended CFGGuidance setting: ComfyUI’s <code>CFG 1.0</code> disables additional classifier-free guidance. In Krea’s official inference code or Diffusers, use <code>CFG 0.0</code> / <code>guidance_scale=0.0</code> for the equivalent setting.Start with LoRA weight <code>1.0</code>. Lower it toward <code>0.85</code> if you want more freedom in clothing, composition, or style. For stronger character influence, experiment with <code>1.1–1.2</code>, checking for exaggerated features or reduced outfit flexibility.For prompting, keep <code>senko_san</code> intact and describe the desired image directly. Detailed natural-language prompts are a strong starting point for Krea 2 Turbo.

Senko-san_NAI-V5-Curated_HD_SUPER (仙狐さん)

NovelAI 5.0 Curated Dataset - 1,000 HD Images

Files with SHA256: fec6a44a99d4e149bb9d5bb3ff4594dfbc453c78eeb813e87b20e6fa3a620799