
模型介绍 - Model Introduction
欢迎使用GuoFeng4模型 - 这是一个微调后的全能的SDXL模型,也可以说是对国人喜欢的画风微调过的模型,具有2.5D,CG,游戏,建模质感。基于SDXL1.0训练。因为SDXL的升级,它的泛化性大大增强,如果你愿意尝试你可以用这个模型画出许多有意思的东西,以及不限于2.5D的画面。
-
Welcome to the GuoFeng4 model - this is a fine-tuned and versatile SDXL model, which can also be said to be a fine-tuned model for the art style that Chinese people like. It has a 2.5D, CG, gaming, and modeling texture. Based on SDXL1.0 training. Due to the upgrade of SDXL, its generalization has been greatly enhanced. If you are willing to try, you can use this model to draw many interesting things, as well as not limited to 2.5D graphics.
======
(经过测试国风4.1几乎不存在这个问题 - After testing, Guofeng 4.1 has almost no such problem)
如果你的出图全身图时出现脸部崩坏建议删除full body关键词或者使用脸部自动修复插件:
国外源地址:https://github.com/ototadana/sd-face-editor.git
国内加速地址:https://jihulab.com/xiaolxl_pub/sd-face-editor.git
-
If you experience facial collapse during the full body image, it is recommended to delete the full body keyword or use the facial automatic repair plugin:
Foreign source address: https://github.com/ototadana/sd-face-editor.git
Domestic acceleration address: https://jihulab.com/xiaolxl_pub/sd-face-editor.git
======
建议参数:
采样器:DPM++ 2M SDE Karras
提示词相关性(CFG Scale):8
Face Editor(或者其他修脸插件):开启
- Suggested parameters:
Sampler: DPM++2M SDE Karras
CFG Scale: 8
Face Editor (or other facial editing plugins): enable
Description
Repeatedly tuning multiple times, including keywords and parameters, this should be the most beautiful 2.5DSDXL model currently available
FAQ
Comments (30)
似乎,好像,又比前两天看上去顺眼点,过两天我再来看看[doge]~
I feel that there may be some problems in the current training, and many tags cannot come out of the prompt
1.0beta画出来是真人感,同样的prompt 1.1画出来只有CG感。
fingers, hands is a very difficult problem, no matter what you do: adetailed , controlnet is faulty. can't make them beautiful and correct
这个模型放目录里之后,选不起是什么原因,老是会跳回上一个使用过的模型上
用的是秋叶整合的包,能升级的都升了,模型加载不上去,不知道是不是配置不够就不给我加载。如果能加载,出图等十几分钟我还是愿意的。我显卡10606g,16g内存,虚拟内存也开了,是挺古董了,不行的话就只能看看了。
目前最低需要8-9G,你可以通过免费的在线网站先运行一下,我B站视频有介绍
6g显存不要想了
想好好玩sdxl就放弃webUI吧,试试comfyui
建议转comfy ui, 虽然上手很难,但是熟练后真的,又快又好,低显存也可以玩
Appears when running this model RuntimeError: Input type (float) and bias type (struct c10::Half) should be the same,May I ask how to solve it
I haven't encountered the reason for this error before. Perhaps you can try forcing fb16
Couple things to try, make sure you are not using an upscaler besides latent, or is sdxl compatible ( dont know of any) and no 1.5 lora's OR Vae's.
Close the HiRes.Fix and try agian?
我发现毛笔,和写毛笔字的训练素材不够,不能让小姐姐很好的写毛笔字。我尝试用过“Chinese calligraphy brush”,“Chinese ink brush”都不怎么行,“ancient Chinese calligraphy brush”好一些些,但不太满意。尤其是“writing with Chinese calligraphy brush”的姿势超级不对,还是西式的握笔方法,能不能补充这样的素材或者加一个LoRA?
能简单介绍一下微调的参数吗?
光感质感都很棒,而且也可实现较多的提示词表达,但是人物表情控制词如果可以加上就好了,比如伤心哭泣,出汗
还有一些控制景别的提示词的用处不大,比如cowboy shot, medium shot, close_up, full body,或者一些控制动作的词
大佬什么时候出个国风动漫类的xl啊
为什么我95%预览图还是好的,到100%就花屏了?是显存不够吗?
vae is set to "auto" or "none"
应该是你VAE的问题,版本不兼容,我刚刚也跟你一样
RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320) 不会解决
RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x2048 and 768x320) 不会解决

















