Type | |
Stats | 85 4 |
Reviews | (12) |
Published | Nov 3, 2024 |
Base Model | |
Training | Steps: 612 Epochs: 3 |
Trigger Words | Claudia Blonde hair, Red eyes Red Dress, White shirt, Red long sleeves, Corset, White skirt |
Hash | AutoV2 8B11184D08 |
A main character from a pixel-style platformer game that was released in 2012.
Sadly the developer has been inactive since 2018.
There are surprisingly some followers still commenting on their site and even trying to make a mod but that also has been no progress since Jan 2023.
Recommended prompt: Claudia, Red eyes, Blond hair, Red dress, Red Puffy Sleeves, White Shirt, White Long Skirt, Corset
I have been using Lazy's guide for SDXL training config but didn't change any settings except for caption and dim/alpha. Nevertheless, I tried to lover the dimension to reduce the size of LoRA but the result was either unsatisfying or way overtrained.
In the end it was still sitting on Config's default 64 dim which resulted in this 400 MB LoRA. I hope I can find more time to test with more settings.
Pony XL is what any new Stable Diffusion should strived to become, Full of passion, Uncensored.
Below are for SD1.5 V2.0
I've removed 2 images from the dataset as they were too "Out of the norm" and now resulted in only 6 images used to train this LoRA. Luckily there are enough variations to allow diverse poses
Below are for V.1.0
This LoRa is actually quite 'Weak' so if the prompt is just 'Claudia' the result might just be a naked girl with a similar face.
Recommended prompt: Claudia, Red eyes, Blond long hair, Red dress, Red Puffy Sleeves, White Shirt, White Long Skirt, Brown Leather Corset
Even with the prompt, the result is 50-50. Usually, it won't have the correct clothing color.
That might be due to the fact that there are only 2 out of 8 training images wearing cloth (Yes, it was a H-game)
Originally trained with my own mix of 7thAnime - Anything - AbyssOrangeMix.
But somehow AnyHentai just gives a result that is more accurate to the original than the trained model