
This LoRA was trained mostly on my images uploaded to Civitai made with PonyXL and a few images made by others. My favourite PonyXL style is Summer Days V2 and I've tried to recreate it for Flux. It is not a 1 to 1 comparison, I would still use Pony for most cases but this LoRA is also interesting anyway. Flux has greater prompt adherence and also you can combine both models via img2img going from one to another. And obviously this LoRA was trained on SFW images of various girls that I uploaded, so no actual ponies here. In my examples I've added masterpiece digital drawing as I think it helps but LoRA was trained without trigger words.
Known pitfalls of this LoRA:
-
Since it was mostly trained on portraits and cowboy shots it struggles a lot with full body images, making short legs and long arms. It's unfortunate so the only workaround is to lower LoRA strength
-
For some reason it tends to do protruding ears
-
It inherited bad hands although I tried to upload only decent images to Civitai
-
It is not Pony, especially when you want to do some character, that you used to do with Pony
But overall I had a lot of fun testing this LoRA and I hope you'll find it useful as well!
描述:
训练词语:
名称: pony-summer-days.safetensors
大小 (KB): 167939
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success