Images used for training are ai-generated by me. Thus the styles can often look sloppy.Honestly I'm not too happy with any of those, but low weight can often be good enough to steer the model towards a specific style without destroying its abilities.<hr />For the StrongLine style, something like <code>expressive colored manga style digital illustration with a realistic touch and extreme ink linework</code> can help steer the model towards the intended style.

Style > AgileStyles

Training Settings: batch size 3, 1024px, 50 images per epoch, optimizer is SinkSGD_advNotes: Need to fully rework the dataset, I'm honestly not too satisfied with how it ended looking. But It can be useful for LoRA combinaison, or to generate the intended style with some effort.

Files with SHA256: 29608fbc37a1aac380cb7d517d531a688d899410dc71279242296262247c27a7