
This is my first LoRA I trained with Kohya scripts and Easy Training Scripts by Derrian Distro of a favorite Hatsune Miku variant of mine.
At first I trained it with Anything 4.5 model, 4 epochs and 5 repeats and results weren't bad but I thought I could've used more, so in comes Apyr (which I cannot thank enough), who used 2 epochs with 40 repeats on my dataset consisting of 36 images and it came out better than before, but he used WebUI with NAI model to train it. Still I'm quite proud of how it came out in the end.
Refeer to the trigger tags on the right in order to get Miku generated with her Magical Mirai 2015/Hand in Hand costume, it works pretty well with Anything and AOM2 when I tested it myself.
描述:
I re-trained this LoRA using settings from a rentry called 'LAZY TRAINING GUIDE' and NAI model. I added lots of new images from dataset (this time with 2 repeats) while removing some s***** ones and edited some to remove text and other distracting elements, I also pruned clothing and body tags so that you only need to use two trigger words to generate her each time without having to describe her at all. I lastly resized it with dim at 8 using Derrian Distro's LoRA Resize script from his collection of scripts you can find on his Github.
It now generates her even better than before in greater quality, which I'm severely impressed by it a lot more than the first time I trained it.
训练词语: magical mirai (2015),hand in hand (vocaloid)
名称: Hatsune Miku (Magical Mirai 2015).safetensors
大小 (KB): 9328
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success