Update 12/24/25: Happy Holidays, folks! For Christmas Eve I've bought to you a V2.0 version of this LoRa that has streamlined training data as to be canon accurate as possible! Scroll down below to learn more.<h2 id="differences-between-v1.0-and-v2.0:">Differences Between V1.0 and V2.0:</h2>V1.0 has significantly less training data compared to V2.0 (18 images compared to 27 images.) V2.0 also has more streamlined training data and additional trigger word(s) (Such as "no ears", "miniboy", etc.) so he wouldn't come out looking like a generic Tumblr Sexyman. In V2.0 I also updated my Illustrious to present cleaner images (V9.0 to V16.0.)Both versions were trained using the XL LoRa trainer on Google Colab.Neither versions were tested with style prompts (Such as "By (Artist Name)") or LoRas, so feel free to use whatever you have on-hand.<h2 id="v1.0-instructional-guide:">V1.0 Instructional Guide:</h2>(Taken directly from V1.0's description.)Trigger word(s) are <code>BS_Spamton, black eyes, colored skin, white skin, long nose, pointy nose, blush stickers, no ears, black hair, hair slicked back, doll joints</code>Default Outfit:<code>white shirt, collared shirt, yellow necktie, red suit, long sleeves, white pants, white shoes</code>Other Trigger Words:<ul><li>(Negative Only) <code>text, watermark</code> - Despite most of the images I used not containing either element, images I generated kept putting in text and watermarks in. I'd put this in the negatives to prevent that.</li></ul><ul><li><code>Closed Eyes</code> - Y'know, because his eyes are depicted as being closed all the time! Still works with the black eyes prompt as well, but you could probably remove it if you don't want to use a bunch of tokens.</li></ul><h2 id="v2.0-instructional-guide:">V2.0 Instructional Guide:</h2>Trigger words are <code>BS_Spamton, colored skin, white skin, no ears, long nose, pointy nose, black eyes, black hair, hair slicked back, white shirt, collared shirt, yellow necktie, red blazer, long sleeves, white pants</code><ul><li>Feel free to also use <code>red pants</code> or <code>black pants</code>. Training data had a rainbow of pants colors, but the most common was the <code>white pants</code> , so I just used that.</li><li>For footwear, I'd recommend just using <code>white shoes</code>. In some images, he's even wearing <code>yellow shoes</code>, so feel free to use that if you prefer it.</li><li>For full body images, use <code>miniboy</code> along with <code>full body</code> to get him to look more canon accurate (Rounded face, short stature, etc.) But if you want him to be a Tumblr Sexyman, feel free to exclude this— But keep in mind I didn't really test it all that much.</li></ul>Other Trigger Words:<ul><li>(Negatives Only) <code>size difference</code> - To avoid having Spamton be tiny in compared to background objects, put "size difference" in the negatives. This will make sure everything is properly proportioned in your images.</li><li>(Negatives Only) <code>text, watermark</code> - If you are paranoid, go ahead and put this in the negatives. Unlike V1.0, watermarks and text appearing in images wasn't a problem, but I noticed in very rare instances it would appear, especially when I was using Regional Prompter. But again, this isn't something to worry about if you don't use Regional Prompt (Like I do.) So feel free to exclude it.</li></ul>One more thing: For some reason, Illustrious V.16 has issues with generating faces on characters, it isn't an issue with the LoRa itself. If this is a recurring issue, I'd recommend just inpainting, most of the images in V.20 were fixed with inpainting and as you can see, it does good. So use it, people.Enjoy using this LoRa!<s>Sorry for all of the NSFW examples in V.20, I just really like Big Shot Era Spamton</s>

Files with SHA256: b8adf8512dee76c3b86453ab9a11dfeab849a69c2d71bb435f46cdc572f7124a