Hello everyone, I decided to train a model at Jougasaki Rika arts. Used NeverEnding Dream (NED) and Anything-V4.5 to create images. Recommendations:<ul><li>Keywords: /RIKA JOUGASAKI/, 1girl, 1girl, solo, long hair, blonde hair, hair ornament, green eyes, two side up</li><li>Use LoRA weight between 0.65-0.85 for versions below 4.0R (there are fewer problems).</li></ul>About versions:Version 4 is now available! Perhaps this is the last version of this character.<ul><li>128 images in various formats and resolutions.</li><li>Fixed issue with teeth.</li><li>Simplified input of tags to call. It is enough to enter 2 or one keyword.</li><li>Like the experimental one, this is a mix of realistic, watercolor, etc. images. More poses like "full body" etc.</li><li>Most of the example images were created using LoRA weight = 1. But in some cases this does not give the best result.</li></ul>About versions 1 and 2:<ul><li>The only difference between versions 1 and 2 is the number of epochs. v.1 - 10 x 55 steps, v.2 - 6 x 55 steps. I'll be happy to read your opinion on this.</li></ul>Version 3 model trained on art with a resolution of 768x768. 🎉🎉🎉<ul><li>Version 4 artmix. (maybe this is not the best idea) 0w0</li></ul>If you have any tips for training LoRA models, I will be glad to read.

(城ヶ崎 莉嘉) RIKA JOUGASAKI

About version:<ul><li>Number of images 125.</li></ul>What does the "V" in the title mean?<ul><li>Used VAE during training.</li></ul>

Files with SHA256: 2ba5a7c8a37719752ba399c2c33788018ff120d7478ec97f903c76eee29df838