Type | |
Stats | 139 202 |
Reviews | (27) |
Published | Oct 7, 2024 |
Base Model | |
Training | Steps: 540 Epochs: 10 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | CAMELLYA_(FORTE) |
Hash | AutoV2 5A1059ED27 |
CHARACTER TABLE
今汐 (Jinshi)
FAQ
Models are too big!
Yes, so it is recommended to use an online platform
The 64-channel pony model is very large, but it can better support multiple skins
Does the character only need one label?
Yes, it is recommended to use a single role tag. Multiple tags means the semantics are not properly separated and absorbed.
Can the character change clothes?
Yes, the character can change clothes
Generally there is a label of XX Regular Clothes to absorb regular clothes
If the character has skin, there will be additional labels to absorb.
Does the character support NSFW?
Yes, the nude data accounts for about 25%-40% of the dataset,
What is the best parameter setting?
There is no best parameter, nor recommended parameter.
In fact, each basic model has its own G-spot, which needs to be explored slowly.
The character does not look like the character in the game
It is possible, because the training mainly uses the masterpieces on pixiv.
Why is the back not lit???
Well, this is indeed a troublesome problem.
There are indeed such pictures in the data set, but Pony seems to have a hard time learning.
The character label is completely invalid
Okay, I may have uploaded the wrong model, thank you for pointing it out, I will replace the correct model.
When to train missing characters
Until I collect more than 20 high-quality works,
Computing is not a problem for me, but good data is hard to collect