Will experiment a bit, this time its trained without captions, only trigger word is used

but the results seems to be good as well. Above 3000 steps is recommended during

training. You can use joycaption to generate the prompt you want and combine together.

Using only the triggerword and caption is a bit weak. I will train another one fully captioned to see what is better.

If you like my work, please consider supporting me: https://ko-fi.com/suzushi2024

I am open to ideas and suggestions on characters lora, models etc.

Also built a free site to generate image and host models, more models are coming soon,

Try it out on : https://miyukiai.com/