大概率是最后一个sdxl模型，经过很多天的测试，不得不承认sdxl是一个在人物方面，非常非常差的模型，甚至比不过sd1.0，人物脸型记忆差，泛化行差，人物没有皮肤纹理，磨皮严重，人物手脚没有任何进步，人体解刨学结构一团糟，人体下半身的表现也是灾难级，如果想用sdxl训练人物的，可以早早放弃，不要浪费时间。训练学习率高，训练时间是sd1.5的10倍，设备要求非常高，没有好显卡基本用起来很困难，目前认为sdxl是在人物训练方面，一个非常失败的大模型。Probably the last SDxl model, after many days of testing,I have to admit that SDxl is a very, very poor model in terms of characters, even inferior to SD1.0. The character's facial memory is poor, the generalization line is poor, the character has no skin texture, the skin is severely worn, the character's hands and feet have not made any progress, the human anatomy structure is messy, and the performance of the lower part of the human body is also catastrophic. If you want to train characters with SDxl, you can give up early and not waste time.The training Learning rate is high, and the training time is 10 times that of SD1.5. The equipment requirements are very high, and it is difficult to use without a good graphics card. At present, it is believed that SDxl is a very failed model in character training.1.这是<a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-0.9/blob/main/sd_xl_base_0.9.safetensors">sdxl-1.0模型</a>的lora测试版本，可能会有很多bug。2.更多测试版本和sdxl 1.0lora交流可以加入tg群：<a target="_blank" rel="ugc" href="https://t.me/sdxlai">@sdxlai</a>3.sdxl lora的使用需要<a target="_blank" rel="ugc" href="https://github.com/AUTOMATIC1111/stable-diffusion-webui/tree/dev">sd dve分支</a>， 起始分辨率1024x1024起步。建议2k，4k分辨率出图4.问题：手脚问题任然存在，无法使用control插件，sdxl的强大clip控制，让关键词生图成为历史。This is the Lora test version of the SDxl-1.0 model, which may have many bugs.2. For more test versions and SDxl 1.0lora communication, you can join the tg group: @ sdxlai3. The use of sdxl lora requires an SD dve branch with a starting resolution of 1024x1024.<h1 id="heading-3">how to use my lora:<a target="_blank" rel="ugc" href="https://civarchive.com/models/46957/how-to-use-my-lora">here</a></h1>Lora model customization: <a target="_blank" rel="ugc" href="https://m.tb.cn/h.Uthb5YE">click here</a>Animation series: <a target="_blank" rel="ugc" href="https://afdian.net/a/wzgrx">Click here</a>Version DescriptionV100: Can change clothes, hair color, eye color, support various main models, support 512-1280 resolutionThe model is made based on <a target="_blank" rel="ugc" href="https://civarchive.com/models/6424/chilloutmix">ch </a>modal。The image source and training code come from the Internet, and the model is only used for exchange of scientific research interests. If there is any infringement, please contact to delete, thank you.图源和训练代码来源于网络，模型仅作为科研兴趣交流使用，如有侵权请联系删除，谢谢！画像ソースとトレーニング コードはインターネットから取得され、モデルは科学研究の利益の交換にのみ使用されます.侵害がある場合は、連絡して削除してください。<h1 id="heading-4">Legal &amp; Risk</h1>It is prohibited to use this model for commercial purposes and any scenarios of illegal acts and purposes.本人原则上不支持使用此模型生成任何成人内容。用户在使用本模型时，请务必遵守当地法律法规，勿侵犯他人的名誉权、隐私权或肖像权

Files with SHA256: c163c5bc5ab9d17d0424414df8b173c3f5dfe999ef572760f0626ab011d82140