<h3 id="first-let's-take-a-look-at-the-instructions-wx6768ooy">First, let's take a look at the instructions</h3>MYH-1.3-GGUF is a small volume model based on MYH-1.3 quantized GGUF-Q4-0, which has faster inference speed and can run with a minimum of 8g of VRAMMYH-1.3 is a model based on MYH-1.1 version (1.1 is based on dev model db training), which is completely fine tuned by dreambooth trainingThis version recommends using Comfyui for optimal performanceThis version requires self loading of clip, t5xxl, and vae during runtimeAdding the use of longclip nodes can achieve better results.<ul><li>I have exiled the accompanying work to the attachment "Training Data " on the right</li></ul>Goose flock:Moyou Alchemy Pavilion - Taiqing Pavilion: 858495398Moyou Alchemy Pavilion - Shangqing Pavilion: 764481772Please also pay more attention to my Little Red Book and Tiktok account: yy469370733<ul><li>模型用途声明</li></ul>1. 你不得将此模型及其衍生版本（如融合模型版本）托管于计划赚取收入或捐赠的网站/应用程序。如果您需要将此模型及其衍生版本用于商业目的（如生成式服务、售卖图片、将图片用于公开发表到短视频平台的文章或出版物等等），请通过站内联系本人2. 您不得直接售卖由此模型生成的图片，除非您对该图片进行了足够程度的人工修改，使其在法律意义上可以被完全判定为您的个人作品。如果您违反本条，所造成的一切法律后果由您个人承担，请恕本人概不负责。3. 您不能使用该模型故意制作或共享非法或有害的内容传播和输出，请您遵守公序良德，将此模型用于积极正面的用途。4. 为了避免不必要的误会，请勿将本模型上传到其他任何模型网站，请勿使用本模型产出的图片训练新模型，并禁止一切形式的融合模型，谢谢！<ul><li>Model Usage Declaration</li></ul>1. You are not allowed to host this model and its derivative versions (such as fusion model versions) on websites/applications that plan to earn income or donate. If you need to use this model and its derivative versions for commercial purposes (such as generative services, selling images, using images for publishing articles or publications on short video platforms, etc.), please contact me through the website2. You are not allowed to directly sell the images generated by this model unless you have manually modified the images to a sufficient extent that they can be fully recognized as your personal work in legal terms. If you violate this clause, all legal consequences caused shall be borne by you personally, and I am not responsible.3. You cannot intentionally create or share illegal or harmful content dissemination and output using this model. Please abide by public order and good morals, and use this model for positive purposes.4. To avoid unnecessary misunderstandings, please do not upload this model to any other model website, do not use the images generated by this model to train new models, and prohibit all forms of fusion models. Thank you!

MYHuman-墨幽随拍-Flux

First, let's take a look at the instructions (click on 'Expand All' to expand all the instructions)1.1：This version is a dreambooth full fine-tuning model, which is a biased experimental model and has not shown any degradation performance in testing. The hand performance should be basically the same as the original dev.The difference from 1.0: 1. Hands are better than v1, and when used alone, it should not be as user-friendly as v1 overall.Can be paired with the MYH-1.1 version of Moyou-F.1-Lora-NetMap to achieve optimized hand capture effects.Lazy people can directly use the v1 instant version.2. V1 is a self training lora merge dev, resulting in severe hand degradation and compatibility degradation with lora. Due to different keys, it is not possible to use Akiba and Azure Dragon scripts as base models to train lora, and db as base models to continue training (this is a common drawback of all lora merge models that use common fusion methods).1.1 is a dreambooth fully fine tuned model with dev as the base model, and no degradation has been found so far. It can be further trained or used as a base model to train lora. It also shows good compatibility with flux dev trained lora (realistic classes may require appropriate weight reduction), and the lora effect trained with itself as the base model is also comparable to that trained with dev as the base model.3. Compatible with hyper-8 step acceleration model (click to download)Quick to get started: (Using fp8 mode locally, the method is the same as using video memory and dev-fp8)Hand gesture: This is a real high-definition photo, (add other after hand gesture)Can be paired with the MYH-1.1 version of Moyou-F.1-Lora-NetMap to achieve optimized hand capture effects.Parameters: Both online and local (webui forge, comfyui) are default.webui-forge：Sampling method (Sampler): Euler or Flux RealisticSchedule type: SimpleIteration Steps: 20Distilled CFG Scale：3.5Guidance coefficient for prompt words (CFG Scale): 1Goose flock:Moyou Alchemy Pavilion - Taiqing Pavilion: 858495398Moyou Alchemy Pavilion - Shangqing Pavilion: 764481772Please also pay more attention to my Little Red Book and Tiktok account: yy469370733

myhumanFlux_myh11.safetensors

MYH-1.1

MYH-1.3-GGUF

MYH-1.3-fp8

MYH-1.3

1.2-train

Ink and Solitude - Flux

先看说明（点击全部展开以展开全部说明）1.1：（使用方法可直接下拉至快速上手）该版本为dreambooth全量微调模型，是一个偏实验的模型，目前没有测试出劣化表现，手的表现应该基本与原dev持平。与v1的区别：1、手比v1好，但整体应该是不如v1好用。 &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;2、v1为自训练lora merge dev，所以手劣化严重，对lora兼容性劣化严重，由于key不同无法使用秋叶及青龙脚本作为底模训练lora，无法db作为底模继续训练（这是所有使用常见的融合方式融合的lora-merge模型共同弊端）。 &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 1.1为dev为底模的dreambooth全量微调模型，目前还未发现劣化，可继续训练也可做为底模训练lora，对flux-dev训练的lora也表现出良好的兼容性(写实类的可能需要适当降低权重)，并且以自身为底模训练的lora效果也与以dev为底模训练的lora效果持平。 &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;3、兼容 <a href="https://www.liblib.art/modelinfo/2b4820cec2e34dba8e9214715d29dd3b?from=personal_page" target="_blank">hyper-8步加速模型(点击下载) </a> 以下是两种采样下的标准出图和8步出图对照测试：<img src="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/2c8ecb52fb55d0061ccb12eb440f1cd1ffbe7c086755926cb03775bea3b5d3a7.jpg?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/2c8ecb52fb55d0061ccb12eb440f1cd1ffbe7c086755926cb03775bea3b5d3a7.jpg?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""> 以下是对以dev为底模训练的lora的兼容性测试：<img src="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/ef2c7eee254afcc37b585a6136acfef735bf3c691811187e5c693d83eaa97926.jpg?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/ef2c7eee254afcc37b585a6136acfef735bf3c691811187e5c693d83eaa97926.jpg?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style="width: 50%;"> 以下是以MYH.1.1为底模训练的lora的兼容性测试：<img src="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/99d321d9f0b478ce0e86e5603726b093057c692049e6b8942e5ca88c5e5292c6.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/99d321d9f0b478ce0e86e5603726b093057c692049e6b8942e5ca88c5e5292c6.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style="width: 50%;"> 快速上手：（在本地使用fp8模式，方法与显存占用和dev-fp8相同）起手式：This is a real high-definition photo,（在起手式后添加其他） &nbsp; &nbsp; &nbsp; &nbsp; 正面tag与 <a href="https://www.liblib.art/modelinfo/b7221470995044809c901a806abf0ec0?msgtype=8" target="_blank">墨幽胶片-Flux</a>、<a href="https://www.liblib.art/modelinfo/dcbc28ab58474fa9a7dda28adffb3d8e?from=personal_page" target="_blank">墨幽3D-Flux</a>、<a href="https://www.liblib.art/modelinfo/e6c19f54cd164687a7384548742ad20b?from=personal_page" target="_blank">墨幽人造人XL</a> 通用，可前往参考例图及返图区tag。参数：线上、本地(webui-forge、comfyui)均为默认。webui-forge：采样方法 (Sampler)：euler 或 Flux Realistic调度类型 (Schedule type)：Simple迭代步数 (Steps)：20Distilled CFG Scale：3.5提示词引导系数 (CFG Scale)：1本地安装： &nbsp; &nbsp; &nbsp;Comfyui：本模型下载后需放入comfyui下models\unet文件夹，若缺少其他依赖模型（clip、vae）可前往 <a href="https://www.liblib.art/modelinfo/2b4820cec2e34dba8e9214715d29dd3b?from=personal_page" target="_blank">Flux-dev/fp8/NF4/GGUF-所需模型文件包</a> 下载并按说明放入指定路径。工作流可前往 <a href="https://www.liblib.art/modelinfo/4ed93da4a9ad48de815985cc9bf41906?from=personal_page" target="_blank" style="text-align: start;">墨幽-FLUX+SDXL-refiner生图矩阵</a> 试用。 &nbsp; &nbsp; Webui-forge：（可直接下载例图导入参数）<img src="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/1019367a3c8d3264b1d05887ae539b216ed4078e5ddc9bf6b2be9a1390b0432d.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/3a9473842c64f01162e10c6d4df24638/1019367a3c8d3264b1d05887ae539b216ed4078e5ddc9bf6b2be9a1390b0432d.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style="">模型按上图设置，配套模型缺少的可前往 <a href="https://www.liblib.art/modelinfo/2b4820cec2e34dba8e9214715d29dd3b?from=personal_page" target="_blank">Flux-dev/fp8/NF4/GGUF-所需模型文件包</a> 下载并按说明放入指定路径。 本模型使用 &nbsp;<a href="https://www.liblib.art/modelinfo/a6a6d91261634be8af50dd886e5142d9?from=personal_page" target="_blank">墨幽-GLM-4v-9b本地标签器</a> &nbsp;打标，有兴趣欢迎前往下载试用。 特别鸣谢 <a href="https://www.liblib.art/userpage/edc8e2bab7614ea9a7e8900ae5c01688/publish" target="_blank">沐辰(点此跳转至主页)</a> 帮我做的封面 鹅群：墨幽炼丹阁-太清阁：858495398墨幽炼丹阁-上清阁：764481772也请多多关注我的小红书和抖音账号：yy469370733

Files with SHA256: 7a6bab0088a39b824ad127285316a7289c7fd0c9a328aa5471306c02a0856545