Type | |
Stats | 87 562 |
Reviews | (25) |
Published | Aug 2, 2024 |
Base Model | |
Training | Steps: 2,090 Epochs: 10 |
Usage Tips | Clip Skip: 1 Strength: 0.6 |
Trigger Words | G4nos bird alien dragonball z muscular red eyes light green skin orange head feathers yellow head feathers 1boy solo +3 more |
Hash | AutoV2 AB2B3F79C6 |
This is a lora of the Dragonball Super villain named Ganos. This one took a number of tries before I realized that the programs thought I was training a Zelda character, on account of Ganos and Ganon being nearly the same name and having the same color pallet theme. Even to the point where it thought I was trying to do a Rito bird character. Funny story and a headache for sure.
Depending on the checkpoint used it works well with a weight ranging from 0.6 to 0.8 with most working well at or near 0.6. Sampling Steps of 30 works best, along with a CFG Scale around 8.
The promts I used to train this lora are: G4nos, bird alien, Dragonball Z, muscular, red eyes, light green skin, orange head feathers, yellow head feathers, 1boy, solo, goggles, goggles, sleeveless shirt, pants.
PS: I changed the word Ganos in my prompts to G4nos so as to avoid making the checkpoints I am training with think I am making a lora of Ganon from Zelda as a rito.
I would advise using words like Zelda, Ganos, and Rito for negative prompts since, even after getting it trained, some checkpoints like to make the picture look close to a Zelda themed picture. I would also add words like Vegita, Goku, balding, high hairline, widow's peak hairline, since for whatever reason the results want to give the character hair that looks like either Vegita or Goku on near most models. I don't know why, and I will chalk it up to the characters probably being popular to use when training checkpoints since they are rather popular characters from a popular anime series. It especially likes to give Vegita's hairstyle the most.