
由于我闲不住,用我空闲的a100显卡随便的训练了这个模型
它的数据集与R-Anime-Sytle-Add-prune_LoRA - v1.0 | Stable Diffusion LoRA | Civitai 一样并增加了一些角色概念
使用了在r-Anime-Sytle上的三倍训练量(rep*3)
区别是它的basemodel为t-ponynai3 5.5,经过我的实际测试,嗯...在这三倍的训练量下,它的画风依旧没有拟合,
pony真是个天坑模型啊~
不过好消息是它似乎学会了不少的角色
如果你想正常使用它 请去R-Anime-Sytle-Add-prune_LoRA - v1.0 | Stable Diffusion LoRA | Civitai 的模型卡页面下载ymal文件,它拥有模型训练的标注词频,当然你也可以使用十六进制查看器 直接查看模型的训练参数
反正我是懒得为这个训练失败的模型统计词频了
Here's the English translation:
Due to my restlessness, I casually trained this model using my spare A100 GPU.
Its dataset is the same as "R-Anime-Sytle-Add-prune_LoRA - v1.0 | Stable Diffusion LoRA | Civitai" with some additional character concepts added.
I used three times the training amount (rep*3) compared to r-Anime-Style.
The difference is that its base model is t-ponynai3 5.5. After my actual testing, well... even with this triple training amount, its art style still hasn't converged.
Pony is truly a bottomless pit of a model, huh~
The good news is that it seems to have learned quite a few characters.
If you want to use it normally, please go to the model card page of "R-Anime-Sytle-Add-prune_LoRA - v1.0 | Stable Diffusion LoRA | Civitai" to download the YAML file. It has the annotation word frequencies from the model training. Of course, you can also use a hexadecimal viewer to directly check the model's training parameters.
Anyway, I'm too lazy to compile word frequencies for this failed training model.
描述:
训练词语:
名称: tpn3-5.5-R-Sytle-400artist-000001 (1).safetensors
大小 (KB): 900219
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success