If you like my work and would like to leave a review, I mean.. feel free to. I do this because I'm a degenerate but also do this for fun.<hr /><h3 id="this-lora-is-part-of-a-collection!">This LoRA is part of a Collection!</h3>Sango (Pony XL) Sango (Illustrious XL) -- you are here Sango (Pony XL/Illustrious XL + Doujins) soon<hr /><h3 id="lora-information">Lora Information</h3>Sango from Inuyasha: The Final Act. Dataset is mostly from that anime. I've re-used datasets and made an attempt to make Illustrious versions. Two versions right now, likely final. They're actually not too different. SangoIllustriousXL_m - Uses a dataset of 120 images - Slightly unrefined, contains tags like "black hair, brown hair" for most. SangoIllustriousXL_m2 - Uses a dataset of 206 images - Slightly refined. Fixed some tags. - More Epochs, same steps. TE LR of 0.0001 I trained a Prodigy version but to be honest I don't care right now. I'm not a current fan of Illustrious XL - personally out of inexperience. PSA. This is not going to mirror the anime, or have anime proportions. Don't expect it. <h3 id="prompt-usage">Prompt Usage</h3> Trigger Word: "Sango" Face/Appearance: "brown eyes, makeup, bangs, black hair or brown hair" Hairstyle: "long hair, low-tied long hair" or "long hair" or "ponytail" Kimono: "kimono, pink kimono, skirt, green skirt" + "fingerless gloves, sandals" (optional) Bodysuit: "bodysuit, black bodysuit, shoulder pads" + "fingerless gloves" ? (optional) Extras: "sword, scabbard" or "weapon, scabbard" for the sheathed katana / "giant boomerang" for the hiraikotsu / not well tagged or understood at the moment. Sample Prompts<pre><code>Sango, 1girl, solo, smile, brown eyes, makeup, bangs, brown hair, long hair, low-tied long hair, breasts, kimono, pink kimono, skirt, green skirt, fingerless gloves, looking at viewer, waving at viewer, standing, cowboy shot, outdoors, day, village</code></pre><pre><code>Sango, 1girl, solo, smile, brown eyes, makeup, bangs, brown hair, ponytail, breasts, bodysuit, black bodysuit, shoulder pads, looking at viewer, hand on own chest, standing, cowboy shot, outdoors, day, village</code></pre><pre><code>Sango, 1girl, solo, annoyed, brown eyes, makeup, bangs, brown hair, long hair, breasts, bodysuit, black bodysuit, shoulder pads, looking at viewer, crossed arms, standing, cowboy shot, outdoors, day, village</code></pre>Do note these are examples. Be sure to use quality tags, expressions.. probably make use of BREAK if your prompts are over 75 tokens. I'm currently learning about that.<h3 id="tips-recommendations">Tips / Recommendations</h3>I recommend a weight of 1, CFG above 5. I am unsure about 1024x1024 vs upscaling. Use Adetailer. You can probably reference the gallery. At the time of writing it is 20 images strong. Models? I've used three in the gallery, they are <a target="_blank" rel="ugc" href="https://civarchive.com/models/827184?modelVersionId=1761560">WAI-NSFW-Illustrious-SDXL</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/992378?modelVersionId=1111838">Prefectious XL NSFW</a>, and <a target="_blank" rel="ugc" href="https://civarchive.com/models/1224788?modelVersionId=1873831">Prefectious Illustrious XL</a>. Last two are similar. Prefect's style is to "booba-ify" characters. Thick proportions. I have no issue with it personally. WAI is a good solid pick. I haven't tried Hassaku's IL variant, could try that.<hr /><h3 id="enjoy!">Enjoy!</h3>

Sango (Inuyasha: The Final Act) | Illustrious XL | NSFW Support

Trained using Civitai (500 buzz), after using Lora Easy Training Scripts for reference. Settings: Network Rank/Alpha: 32/16 Optimizer: Adafactor, cosine with restarts, 3 restarts Learning Rate: Unet 0.0005, Tenc 0.00005 Others: Noise Offset of 0.1, Multires Noise Iterations of 6, Discount 0.3, SNR Gamma of 5 Dataset: 120 images, 4 epochs, 9 repeats, 4 batch size, 1080 steps + Keep Tokens, Flip Augmentation, Shuffle Captions, No Regularization

Files with SHA256: 1319eb247d93beee8492826d95e11edee7e1da64380ef285e2817f3297d2f8c1