<p>High-fidelity character LoRA for <strong>Senko-san</strong> from <strong>The Helpful Fox Senko-san</strong>, trained for <strong>Anima</strong>.</p><p>Built from a high-quality synthetic dataset of <strong>598 captioned images</strong> generated with <strong>NovelAI 4.5 Curated</strong>. Designed for <code>txt2img</code>, <code>img2img</code>, inpainting, and character replacement workflows.</p><p>The trigger word is:</p><pre><code>senko_san</code></pre><h2 id="what-this-model-does-best">What This Model Does Best</h2><ul><li><p><strong>Character identity:</strong> Reproduces Senko's recognizable face, amber eyes, short layered golden-blonde hair, fox ears, fluffy fox tail, hair ornament, and anime appearance.</p></li><li><p><strong>Outfit flexibility:</strong> Maintains her identity across dresses, casual clothing, jackets, hoodies, kimono, and other outfits.</p></li><li><p><strong>Natural-language prompting:</strong> Works well with descriptive prompts suited to Anima's Qwen3-0.6B text encoder.</p></li><li><p><strong>Scene flexibility:</strong> Handles indoor, outdoor, traditional, modern, daytime, sunset, and nighttime environments.</p></li><li><p><strong>Img2img and inpainting:</strong> Can apply Senko's identity while preserving an existing composition, clothing, pose, lighting, or background.</p></li></ul><h2 id="recommended-generation-settings">Recommended Generation Settings</h2><p>SettingRecommendation<strong>Base model</strong>Anima Base v1.0 or compatible Anima checkpoint<strong>Trigger word</strong><code>senko_san</code><strong>Prompt style</strong>Natural language or mixed natural language + tags<strong>LoRA weight</strong><code>0.85–1.0</code><strong>Sampler</strong>Euler Ancestral (<code>euler_ancestral</code> / <code>Euler a</code>)<strong>Scheduler</strong>Simple<strong>Steps</strong><code>28</code><strong>CFG scale</strong><code>7</code><strong>Resolution</strong><code>1024 × 1024</code> or another suitable Anima resolution</p><p>For stronger character influence, LoRA weight <code>1.2</code> can also work well.</p><h2 id="prompting">Prompting</h2><p>Anima uses <strong>Qwen3-0.6B</strong> for text conditioning, so descriptive natural-language prompts work well.</p><p>Instead of relying only on short tag lists such as:</p><pre><code>senko_san, 1girl, fox ears, blue dress, beach, sunset, sun hat</code></pre><p>you can describe the image more directly:</p><pre><code>senko_san in an anime-styled image. She is standing on a beach at sunset wearing a light blue long dress and a sun hat. The view is from the front, with warm golden-hour lighting and colorful clouds behind her. Her fluffy fox tail hangs naturally to the side.</code></pre><p>Tags and natural language can also be mixed together.</p><p>A short negative prompt is usually a good starting point:</p><pre><code>worst quality, low quality, blurry, deformed, text, watermark</code></pre><p>More targeted negatives can be added when needed for ears, tails, anatomy, or other recurring problems.</p><hr /><h2 id="example-1-beach-sunset">Example 1 — Beach Sunset</h2><p><strong>Prompt:</strong></p><pre><code>senko_san in anime styled image. Breathtaking beach sunset, close-up of senko_san. Golden hour sunset. senko_san wearing a light blue long dress. View is from the front facing senko_san. Her tail is to the side and down. Some clouds in the sky. The sunset sky is awash in various colors. She wears a sun hat. Use senko_san’s recognizable face, short layered hair, fox ears, fluffy fox tail, and proper colors. Hair color: #F2B93B to #F6C84A with highlights around #FFD86A. Fur color: #EFB53A to #F5C246 with soft cream tones and white fur accents. Keep the result warm golden-yellow, not pale and not overly orange.</code></pre><p><strong>Negative prompt:</strong></p><pre><code>{{{worst quality, low quality}}}, {{{multiple tails}}}, {{{cat ears}}}, {{{human ears}}}, child, loli, petite, blurry, deformed, mutated, signature, watermark, text, username, multiple girls, crowd, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, jpeg artifacts, bad feet, extra limbs, mutated hands, fused fingers, too many fingers, malformed limbs, extra limbs, extra ears, bad ears, lowres, artistic error, film grain, scan artifacts, worst quality, bad quality, jpeg artifacts, very displeasing, chromatic aberration, dithering, halftone, screentone, multiple views, logo, too many watermarks, negative space, blank page</code></pre><p>ParameterValueWorkflow<code>txt2img</code>LoRA weight<code>1.2</code>SamplerEuler AncestralSchedulerSimpleSteps<code>28</code>CFG scale<code>7</code>Resolution<code>1024 × 1024</code></p><hr /><h2 id="example-2-traditional-kimono">Example 2 — Traditional Kimono</h2><p><strong>Prompt:</strong></p><pre><code>senko_san in an anime-styled illustration. She is sitting comfortably in a traditional Japanese room while wearing an elegant kimono. She holds a small tea cup in both hands and smiles warmly toward the viewer. Soft afternoon light enters the room and creates a peaceful, welcoming atmosphere. Use senko_san’s recognizable face, amber eyes, short layered golden-blonde hair, fox ears, fluffy fox tail, and hair ornament.</code></pre><p><strong>Negative prompt:</strong></p><pre><code>worst quality, low quality, blurry, multiple tails, cat ears, human ears, deformed anatomy, bad hands, extra fingers, extra limbs, text, signature, watermark</code></pre><p>ParameterValueWorkflow<code>txt2img</code>LoRA weight<code>1.0</code>SamplerEuler AncestralSchedulerSimpleSteps<code>28</code>CFG scale<code>7</code>Resolution<code>1024 × 1024</code></p><hr /><h2 id="example-3-leather-jacket-and-neon-city">Example 3 — Leather Jacket and Neon City</h2><p><strong>Prompt:</strong></p><pre><code>senko_san in an anime-styled nighttime city scene. She is wearing a black leather motorcycle jacket with blue jeans while standing on a neon-lit city street. Bright signs and colored lights reflect from the pavement behind her. She faces the viewer with a friendly expression. Keep senko_san clearly recognizable with her amber eyes, short layered golden-blonde hair, fox ears, fluffy golden fox tail, hair ornament, and warm golden-yellow coloring.</code></pre><p><strong>Negative prompt:</strong></p><pre><code>worst quality, low quality, blurry, multiple tails, cat ears, human ears, multiple girls, crowd, deformed anatomy, bad hands, extra fingers, extra limbs, text, signature, watermark</code></pre><p>ParameterValueWorkflow<code>txt2img</code>LoRA weight<code>0.85</code>SamplerEuler AncestralSchedulerSimpleSteps<code>28</code>CFG scale<code>7</code>Resolution<code>1024 × 1024</code></p><hr /><h2 id="example-4-oversized-hoodie">Example 4 — Oversized Hoodie</h2><p><strong>Prompt:</strong></p><pre><code>senko_san in a cozy anime-styled modern kitchen during the morning. She is wearing an oversized white hoodie and holding a warm coffee mug. Gentle morning sunlight enters through a nearby window and softly illuminates her face and hair. Keep senko_san’s recognizable facial features, amber eyes, short layered golden-blonde hair, fox ears, fluffy fox tail, hair ornament, and proper golden-yellow coloring.</code></pre><p><strong>Negative prompt:</strong></p><pre><code>worst quality, low quality, blurry, multiple tails, cat ears, human ears, shrine maiden outfit, multiple girls, deformed anatomy, bad hands, extra fingers, extra limbs, text, signature, watermark</code></pre><p>ParameterValueWorkflow<code>txt2img</code>LoRA weight<code>0.90</code>SamplerEuler AncestralSchedulerSimpleSteps<code>28</code>CFG scale<code>7</code>Resolution<code>1024 × 1024</code></p><hr /><h2 id="example-5-img2img-inpainting">Example 5 — Img2img / Inpainting</h2><p><strong>Prompt:</strong></p><pre><code>Change the character into senko_san while preserving the original image's pose, camera angle, clothing, lighting, and background composition. Give her senko_san’s recognizable face, amber eyes, short layered golden-blonde hair, fox ears, fluffy golden fox tail, and characteristic hair ornament. Keep the expression natural and make the character blend cleanly with the existing image.</code></pre><p>For face or head replacement:</p><pre><code>Replace the character’s face and head with senko_san. Give her senko_san’s recognizable facial features, amber eyes, short layered golden-blonde hair, fox ears, and hair ornament. Preserve the existing body, clothing, pose, lighting, camera perspective, and background.</code></pre><p><strong>Negative prompt:</strong></p><pre><code>worst quality, low quality, blurry, multiple tails, cat ears, human ears, deformed face, bad anatomy, extra ears, extra limbs, text, signature, watermark</code></pre><p>ParameterValueWorkflow<code>img2img</code> or inpaintingLoRA weight<code>0.85</code>Denoising strength<code>0.55</code>SamplerEuler AncestralSchedulerSimpleSteps<code>28</code>CFG scale<code>7</code>Suggested input size<code>1024 × 1024</code></p><p>Increase denoising toward <code>0.7</code> when a stronger identity change is needed.</p><h1 id="training-parameters">Training Parameters</h1><h2 id="engine-and-base-model">Engine and Base Model</h2><p>ParameterValueTraining engine<code>ostris/ai-toolkit</code>Base modelAnima Base v1.0LoRA typeStandard LoRATrigger word<code>senko_san</code>Training precisionBF16</p><h2 id="training-schedule">Training Schedule</h2><p>ParameterValueTraining steps<code>2,000</code>Batch size<code>2</code>Gradient accumulation<code>1</code>Diffusion model trainingEnabledGradient checkpointingEnabledCheckpoint intervalEvery <code>200</code> steps</p><h2 id="network-architecture">Network Architecture</h2><p>ParameterValueNetwork rank / dimension<code>32</code>Network alpha<code>32</code>Effective alpha/rank scaling<code>1.0</code></p><h2 id="optimizer-and-learning-rate">Optimizer and Learning Rate</h2><p>ParameterValueOptimizerAutomagicDiffusion model learning rate<code>0.0001</code>SchedulerCosine</p><h2 id="dataset-and-resolution">Dataset and Resolution</h2><p>ParameterValueTraining images<code>598</code>Dataset sourceSynthetic images generated with NovelAI 4.5 CuratedCaption format<code>.txt</code> captionsBase resolution<code>1024 × 1024</code>Caption shufflingEnabledPreserve Tag<code>1</code>Caption dropout<code>0.05</code>Latent cachingEnabled</p><h2 id="text-encoding">Text Encoding</h2><p>Anima uses <strong>Qwen3-0.6B</strong> for text conditioning.</p><p>Natural-language descriptions are recommended for detailed control over clothing, poses, composition, lighting, environments, colors, and character details. Tags can still be used on their own or mixed with descriptive text.</p><p>There is no Clip Skip setting.</p><h2 id="recommended-starting-point">Recommended Starting Point</h2><p>For general use:</p><pre><code>LoRA: 0.85–1.0
Sampler: Euler Ancestral
Scheduler: Simple
Steps: 28
CFG: 7
Resolution: 1024 × 1024</code></pre><p></p>

Senko-san_NAI-V4.5-Curated

<p><strong>NovelAI 4.5 Curated</strong> Dataset - <strong>598</strong> Images</p>

Files with SHA256: 0b577de0843d98e350e0c5573b5429b4faf5c0e8f54533dfc8bfd4078b38678d