这是我自己使用的工作流:https://civarchive.com/models/2986628/qwen-image-21-editflow-lora
This is the workflow I use: https://civarchive.com/models/2986628/qwen-image-21-editflow-lora
V3:更稳定的控制姿势 CFG建议设置为3,高CFG可以让qwen image2.1的出图效果更稳定
V3: More stable pose control. A CFG value of 3 is recommended. Higher CFG values can make Qwen Image 2.1 generation more stable.
V2:新增阴毛(需自行描述,默认无),尽量保护原有饰品和发型
V2:Add pubic hair (describe it manually; none by default),Preserve the original accessories and hairstyle as much as possible..
# Qwen-Image-2.1 NSFW Image Edit LoRA
全裸编辑 / Nude Edit · 开腿 / Legs Spread · 女上位 / Cowgirl · 口交 / Oral
给 Qwen-Image-2.1 图像编辑 用的 LoRA。
输入一张原图 + 一句中文指令,把着衣照改成裸体,或改成开腿、女上位、口交,尽量保住脸、发型、饰品和原构图。
成人向。训练目标是 undress to nude,不是普通换装。
## 能做什么
- 全裸 / Nude:脱掉指定衣服,全身裸露或只脱上身;袜子、鞋子、首饰、手机、伞、发型、表情、姿势可按指令保留。
# Qwen-Image-2.1 NSFW Image Edit LoRA
Nude Edit / 全裸编辑 · Legs Spread / 开腿 · Cowgirl / 女上位 · Oral / 口交
A LoRA for Qwen-Image-2.1 image editing.
Give it a source image plus one Chinese instruction: turn a clothed photo into a nude, or into a legs-spread / cowgirl / oral edit, while keeping the face, hairstyle, accessories, and original composition as much as possible.
NSFW. Trained for undress to nude, not for normal outfit swap.
## What it does
- Nude / 全裸: Remove the listed clothes. Full nude or upper-body only. Socks, shoes, jewelry, phone, umbrella, hair, expression, and pose can be kept if you say so.
## How to write the prompt
Use Chinese. The training captions are all short Chinese edit instructions. English prompts were barely trained.
Description
新增阴毛,需要自己描述,默认为没有
FAQ
Comments (16)
你的毛真贵~!!
I guess adding pubes to hide deformed pussy is not that bad idea LOL|
But where did you even collect a dataset like that these days? LOL
does v1 tend to soften the skin? idk but the other nsfw checkpoint doesn't do that but the genitals are just horror when generated
Bro why are you keeping too much buzz. 200 was alright. Its a new model atleast let the community grow first.
I’ll remove the paywall in a couple of days, so just hang tight. I originally set it as early access, but for some reason it ended up becoming permanently paid.
用下来感觉效果仍然有待提高,还不能替代已有的方案
问题主要表现为:很多情况下乳头和私处会丢失,不过就Lora成功工作的案例来说,效果很棒。
还有的问题是,当我加载lora后,似乎对其他动作无法做出响应,感觉可能是模型本身的问题
如果后续可以完美实现预期效果的话,对于7B的模型,这将会令人愉快
我倒是还没遇到确实乳头和私处的,可能我本身都是权重测试的,动作无法响应的话我只试过一些简单的动作,这两个都是rank16训练的,后面我试试提高rank能否提高强度
大佬,我是新人,有什么工作流可以推荐一下吗
用官流,或者B站搜黑鹤佬的视频,他的工作流都挺好用的
给你这个,我自己设置的工作流,文/图通用。很好用。设置基本不用再改了。只接入了6张图片应该够用。
可以改的内容:LoRA、加速节点(可忽略,加速运行速度快,但有损些许画质。)、图片2-6的图像缩放(缩放运行速度快,但有损。当质量优先的情况下,可以关闭缩放。原图输入。)、提示词和图片的种子、百万像素、宽高比。
工作流:https://gofile.io/d/gGcNkJZa
所要用到的插件:https://github.com/xiaowuapple-pixel/ComfyUI-Prompt-Enhancer
https://github.com/goohai/Goohaitools-comfyui
https://github.com/tl2012tl/TE_MAN
https://github.com/tl2012tl/TE-Speed-QwenImage21
https://github.com/rgthree/rgthree-comfy
https://github.com/LAOGOU-666/Comfyui-Memory_Cleanup
所要用到的额外模型(放置到 LLM文件夹):https://huggingface.co/pottokao/Qwen-Image-2.1-PE-T2I-Heretic-GGUF/resolve/main/pe_t2i_heretic-Q8_0.gguf
@hoyjooh170 你分享的工作流链接打开后显示“此内容不公开可访问,要使此内容公开可访问,其所有者必须更改其可见性设置。”
@ui_90 现在可以了。前面没弄好。 打开图片输入就是图生图,关闭所有图片输入节点则是文生图。另外我再次修改了一下工作流,现在更合理了。
太棒了,我用了这个lora才成功破甲
效果不错,不过相对来说这个LoRA有一定的磨皮、美化效果。用了它和不用它比起来,有大概率脱了之后皮肤更白、身材更好。








