禁止将模型以任何形式进行商用<h1 id="v3.0">v3.0更新内容：</h1>1、修复了人物的呆毛2、修复了人物的发卡图案（必须开启ADetailer才能修复，）（当然侧身的时候概率无法修复）（无法修复可以选择删除tag:hairclip）3、人物的吊坠勉强能不糊（特写镜头太少了）4、人物的头发概率能还原蓝色的一小撮头发5、这版不建议出现全身图，为了发卡图案，只能牺牲全身图，鞋子部分出现的数量很少。6、发卡和吊坠可移除（吊坠概率可移除，发卡删除tag：hairclip，吊坠删除tag：pendant,jewelry）如果无法删除，尝试放入负面提示词中。如果无法修复那只能说免费额度不足以支撑我使用更多的图片，70张已经是免费的极限了，虽然可以通过降低轮数来解决，但实际上也多不了几张图片，同时免费colab无法使用bf16只能这样了，很抱歉我的电脑无法进行模型的训练，显存仅有8GB，勉强只能达到跑图需求，远达不到训练需求。如果电脑好点就好了，直接用几百张数据集肯定能完美修复吊坠和发卡问题。就这样吧，v5.0 Update Notes:1. Fixed character Ahoge2. Fixed character hairclip patterns (requires ADetailer to be enabled for repair) (Note: repair probability remains low during side-view animations) (If repair fails, consider deleting tag:hairclip)3. Character pendants now display with minimal blurring (close-up shots remain scarce)4. Character hair may occasionally restore a small tuft of blue hair5. Full-body shots are not recommended in this version; full-body shots must be sacrificed for hair clip patterns. Footwear appears very rarely.6. Hairpins and pendants are removable (pendants may be removed probabilistically; hairpins require deleting tag:hairclip, pendants require deleting tag:pendant,jewelry). If deletion fails, attempt adding negative prompt words.If the issue persists, it must be acknowledged that the free quota is insufficient to process additional images. Seventy images represent the absolute limit of the free tier. While reducing the number of iterations could offer a workaround, this would yield only marginally more images. Furthermore, the free Colab instance cannot utilise BF16, leaving this as the only viable solution. Apologies that my personal computer cannot support model training; with merely 8GB of VRAM, it barely meets the requirements for image generation, let alone training. If only my computer were better—a dataset of several hundred images would undoubtedly fix the pendant and hairclip issues perfectly. Well, that's how it is.<h3 id="v2.0ps">v2.0更新，修改了数据集，但发卡和吊坠变糊了，力竭了，明明用ps处理了还是没办法。</h3>（已深度思考，用时0秒）喵！（已深度思考，用时0秒）喵！（已深度思考，用时0秒）喵！（已深度思考，用时0秒）喵！（已深度思考，用时0秒）哈！（已深度思考，用时0秒）哈！（已深度思考，用时0秒）哈！<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/12ac4f71-4b06-4e7e-9df6-cf3f51e54949/original=true/12ac4f71-4b06-4e7e-9df6-cf3f51e54949.jpeg" />处理图片有点过度，导致人物头发变色了，衣服也挺糊的，应该不会改了，除非没人做。实际上是等了两天没人炼，我很失望，自己动手先满足一下，懒得修改了，等别人炼个更好的，我直接选择白嫖。此模型为代餐，等别人炼个更好的。The image processing was a tad excessive, resulting in discoloured hair and rather muddy clothing. I doubt it'll be revised unless someone else takes it on. Truth be told, I waited two days with no one refining it. Disappointed, I decided to satisfy myself first. Can't be bothered to tweak it now—I'll just wait for someone else to refine a better version and then leech off their work. This model is merely a stopgap until someone else refines a superior one.上传的是10轮的结果，图片是第5轮的结果The uploaded file contains the results of 10th Epochs, with the image showing the results of the 5th Epochs.因为是测试模型，所以这次就不发tag了，可以参考图片。虽然图片中没有腿部的tag，毕竟全身图真的很糊。Since this is a test model, I won't be adding tags this time—please refer to the images instead. Though the leg tags aren't visible in the pictures, the full-body shots are quite blurry after all.

Files with SHA256: 5cb8fef49b50823b38a24182d5cd577d37c33fc2d2e33efdc689e120c086f282