Type | |
Stats | 154 178 |
Reviews | (20) |
Published | Oct 11, 2024 |
Base Model | |
Training | Steps: 640 Epochs: 10 |
Usage Tips | Clip Skip: 2 |
Trigger Words | ningning tank top harness cutout strapless camisole jewelry necklace straps skirt rolled-up and knotted skirt +7 more |
Hash | AutoV2 E21E66D8CC |
Cloth Parted Supported!!
That means you can trigger the boots only if you like.
====================================================
I wanted to make a lora about this costume's boots long time ago,
but with rare dataset of this costume(yes,there are only three video cuts and 2pics on internet, none fan shots, none official stage shots, none news shots.) and my indolence.
I give up.
But recently, I was inspired by @AIadoreyou
With her kindness and enthusiasm, I decide to give another try.
So I experiment a new way collecting datasets which inspired by @PPFemdom
I think he cut video into several clips, and take the stable & clean pictures as dataset.
So I clip the whole 3:24 time long video into 2.79Gb pictures, I only picked about 10 pictures, cause I think the rest of them are almost the same angle (facing to viewer). Then I search the same type cloth、skirt、necklace、boots on online shop. There are many good clean shots, and no worry if someone is wearing it, I only use it for still life and detail learning.
I tried twice, the second version seems okay.
but the tagging part still needs improve I think.