This is an anime model that I made to try to match or exceed the effect of the novel ai v3 (obviously I failed), but in the end got a better result, so I released this version.

Precautions for use:

Baked vae

Steps 30

Sampling method DPM+ 2M SDE Heun Exponential

CFG 8-20 CFG Scale Fix is required if high CFG is used

Negative tag: bad quality,bad anatomy,worst quality,low quality,lowres,extra fingers,blur,blurry,ugly,wrong proportions,watermark,image artifacts,

Please open HD repair

I am personally opposed to using open source models and images to train closed-source commercial behavior like novel ai and midjourney, but it has been a long time since SDXL was released and I have to admit the awkward fact that there is no model in the community that can do as well as those commercial models, which is an embarrassing fact. Maybe one person can't do it.

Regardless of whether the novel ai model will leak, I think we need a method rather than the model itself.

I will try to steal as much information as I can to help the open source community, but I also need help, if someone can steal the training logs of novelai please contact me, I will rent the graphics card build data set and try to train better models to give back to the community.

这是一个动漫模型，我制作它旨在尝试达到或超过novel ai v3的效果（显然我失败了），但终归是得到了一个比较好的结果，所以我发布了这个版本。

使用注意事项：

以烘焙vae

步数 30

采样方法 DPM++ 2M SDE Heun Exponential

CFG 8-20 如果用高cfg需要CFG Scale Fix

负面tag： bad quality,bad anatomy,worst quality,low quality,lowres,extra fingers,blur,blurry,ugly,wrong proportions,watermark,image artifacts,

请打开高清修复

我本人很反对使用开源的模型和图片训练后闭源商用的行为就像novel ai和midjourney,但是自从SDXL发布后已经很久了不得不承认一个尴尬的事实，社区里目前没有模型可以做到媲美那些商用模型，这是是个很尴尬的事实，也许一个人的力量并没法做到。

先不提novel ai的模型是否还会泄露，我觉得我们需要的是一个方法而不是模型本身。

我会倾尽所能尝试偷取到更多的信息来帮助开源社区，同时我也需要帮助，如果有人能偷取novelai的训练日志请联系我，我会资费租用显卡的构建数据集尝试训练出更加优异的模型回馈社区。