Basically I made a lora inspired by 2 drawings and a comic from supersatanson mewtwo just to see what is the minimum for a dataset
The first version was made with 4 images and is really strong so you should used it at .4 or less if you dont want the original image
Then second version was made with the same 4 images but I increase the dataset using the first version, it works better at .6 to .8 but results can vary greatly
If you have any suggestions for a character or style let me know in the comments.
Español
Me gusta el artista supersatanson e intente hacer un lora de un personaje que tiene 2 dibujos y un comic para ver cual es el minimo para un dataset
La primer version en niveles por encima del .4 hace que genere las imagenes originales pero me sirvieron para incrementar la muestra para del dataset de la siguiente version
Si tienen cualquier sugerencia para algun personaje aunque tenga solo una imagen podria intentar hacerlo
描述:
The first version was made with 4 images and is really stron so you should used it at .4 or less if you dont want the original image
训练词语: mewtwo
名称: princess_mewtwo-10.safetensors
大小 (KB): 18545
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
