<h3 id="important:-read-below-eu0atl3h6"><a target="_blank" rel="ugc" href="https://ko-fi.com/ctian"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/cf614846-63c8-4596-a5fd-f5962d69ae75/width=525/cf614846-63c8-4596-a5fd-f5962d69ae75.jpeg" /></a>IMPORTANT: READ BELOW</h3><div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/ZZvIVRQ4E7I" start="0"></iframe></div>Trigger word(s): ctiank0shi, blonde hair, Clothing: serafuku, blue sailor collar, blue neckerchief, puffy long sleeves, miniskirt, Example Positive prompt ✅:<pre><code>score_9, score_8_up, score_7_up, source_anime, masterpiece, 1girl, ctiank0shi, serafuku, blue sailor collar, blue neckerchief, puffy long sleeves, miniskirt, looking at viewer, outdoors, city, sunny, depth of field, day, shiny skin, cowboy shot, sitting, legs,</code></pre>Negative prompt ❌:<pre><code>sketch, 3d, claws,</code></pre>Recommended resolution, aspect ratio for sdxl, pony 📷768 x 1344: Vertical (9:16)832 x 1216: Portrait (4:5) **1024 x 1024: Square (1:1)1182 x 886: Photo (4:3)1254 x 836: Landscape (3:2)1365 x 768: Widescreen (16:9)Work flow, procedure 🔧1. Collect dataset (booru, anime screencaps, search)2. Clean up dataset using upscalers, inpainting and photoshop3. Tag dataset using booru <a target="_blank" rel="ugc" href="https://github.com/starik222/BooruDatasetTagManager">github</a>4. Train LoRa in kohya ss locally <a target="_blank" rel="ugc" href="https://github.com/bmaltais/kohya_ss">github</a>5. Test epochs using xyz grid in controlnet to ensure LoRa is not overfitted/underfitted.6. Submit model to CivitAI <a target="_blank" rel="ugc" href="https://civarchive.com/user/ctian">profile</a>do not ask me training parameters, check lora metadata for training parameters<h1 id="faq:-1csb0klrc">FAQ:</h1>Q: why not use character name as trigger word? A: I use unique token to prevent other things from being prompted when using triggerword Q: do you have other account in other websites? A: yes, tensor, pixai Q: what external generator do you use? A: a111 mostly, comfyui for img2vid

Koshi Torako 虎視 虎子 PONY - Shikanoko Nokonoko Koshitantan / my deer friend nokotan

35 images added to training data ep1-3 included

Koshi_Torako_Pony_ct_ver2.safetensors

v2.0

v1.0

<h3 id="important:-read-below-yrutohr17"><a target="_blank" rel="ugc" href="https://ko-fi.com/ctian"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/cf614846-63c8-4596-a5fd-f5962d69ae75/width=525/cf614846-63c8-4596-a5fd-f5962d69ae75.jpeg" /></a>IMPORTANT: READ BELOW</h3><div data-youtube-video><iframe width="640" height="480" allowfullscreen="true" autoplay="false" disablekbcontrols="false" enableiframeapi="false" endtime="0" ivloadpolicy="0" loop="false" modestbranding="false" origin playlist src="https://www.youtube.com/embed/ZZvIVRQ4E7I" start="0"></iframe></div>Trigger word(s): ctiank0shi, blonde hair Clothing: serafuku, sailor collar, blue neckerchief, skirt, long sleeves, Example Positive prompt ✅:<pre><code>score_9, score_8_up, score_7_up, source_anime, masterpiece, 1grl, ctiank0shi, blonde hair, serafuku, sailor collar, long sleeves, blue neckerchief, skirt, long sleeves, looking at viewer, outdoors, sunny, city, depth of field, </code></pre>Negative prompt ❌:<pre><code>sketch, 3d, claws,</code></pre>Recommended resolution, aspect ratio for sdxl, pony 📷768 x 1344: Vertical (9:16)832 x 1216: Portrait (4:5) **1024 x 1024: Square (1:1)1182 x 886: Photo (4:3)1254 x 836: Landscape (3:2)1365 x 768: Widescreen (16:9)Work flow, procedure 🔧1. Collect dataset (booru, anime screencaps, search)2. Clean up dataset using upscalers, inpainting and photoshop3. Tag dataset using booru <a target="_blank" rel="ugc" href="https://github.com/starik222/BooruDatasetTagManager">github</a>4. Train LoRa in kohya ss locally <a target="_blank" rel="ugc" href="https://github.com/bmaltais/kohya_ss">github</a>5. Test epochs using xyz grid in controlnet to ensure LoRa is not overfitted/underfitted.6. Submit model to CivitAI <a target="_blank" rel="ugc" href="https://civitai.com/user/ctian">profile</a>do not ask me training parameters, check lora metadata for training parameters<h1 id="faq:-h94x9nr4t">FAQ:</h1>Q: why not use character name as trigger word? A: I use unique token to prevent other things from being prompted when using triggerword Q: do you have other account in other websites? A: yes, tensor, pixai Q: what external generator do you use? A: a111 mostly, comfyui for img2vid

&lt;p&gt;35 images added to training data ep1-3 included&lt;/p&gt;

Files with SHA256: 8d105f1f8861b7b02b0630f05dd5c3fdfbdfeda9fbbb62d54c7ccb4489e115bd