This LoRA was trained with over 100 carefully selected images to replicate Sakimichan's current style as faithfully as possible. Overall, the LoRA achieved very good results, but I am not entirely satisfied with it. In my opinion, it doesn’t quite achieve the balance I’m looking for between realism and the anime style that Sakimichan masters, and some shadows still feel off to me. However, for a first version, I think it’s pretty good. I’ll keep improving the LoRA, and I would greatly appreciate it if you could leave a few comments to help me refine it further.
A major advantage of my model compared to others is that you don’t have to worry about watermarks, as I manually removed them from each image. As a result, the model wasn’t trained with Sakimichan’s signature or Patreon logos.
Recommendations
Model: Flux Base works very well with this LoRA, and it’s the one I recommend for your creations.
Sampler: DPM 2 Ancestral with the Karras scheduler has shown the best performance in my tests.
Sampling Steps: Between 15 and 40 steps produce excellent results.
Scale: I recommend a scale between 8 and 11.