More words have been added to the training data and the colors should now be a bit more vibrant.
If you like the character, always leave a like and a picture, and if you give some buzz you will allow me to train other LORAs or MODELS. Downloading only is not useful. For model visibility, remember to rate the images in my gallery and leave a like. Thank you
Version 2a is a further development of Version 1, but with a bit more freedom in the Design. It moves away somewhat from the Anime Style.
Version 2b is almost identical to Version 2. The difference between the two is that in
Version 2b, Eira Kishida once again uses the Design from the Anime.
Version 3 is an evolution of Version 2B and contains new prompts.
Version 3B is an alternative to Version 3. It is experimental.
Version 4, the trainings Data is bigger (more Pictures and Prompts used)
Version 5, the trainings Data is from Version 4 but the Prompts/Pictures are improved.
Version 6, the trainings Data is improved
Recommended Settings Version 5:
Lora: 0.9-1.0
CFG Scale 3.5-6.0
Steps: 24-50
Sampler: Euler a
Trained with AdamW8Bit
Engine: Kohya
Recommended Settings Version 6:
Lora: 0.9-1.0
CFG Scale 3.5-5.0
Steps: 24-50
Sampler: Euler a, Trained with AdamW8Bit
Engine Kohya