Despite the fact that Lora is trained only on portrait photos, she is capable of much more.
this LORA was directly aimed at improving details in portrait photos (but is capable of more). Much better eyes, hair detail, faces, overall clothing detail, better proportions, better anatomy, very deep and varied style
New version (better natural colors, small details, anatomy & eyes):
Face Detailed v2 (Boosty)
Face Detailed v2 (Patreon)
worlflow
I trained lora on this model RouWei v0.7 (vpred)
if you don't use this model, the quality will be worse!
RouWei is essentially a continuation of training on a large amount of data (more than 7 million images) on Noobai. All the same problems, but even more accurate.
weight - any, lora is set ideally to 1, you can reduce it
This lora can reduce the contrast of any v-pred model, you can check.
can I use non booru tags? Yes, but it is better to leave the tagging system, an example of partial use of booru tags and random made-up tags with a detailed description (don't forget the initial tags masterpiece, best quality, newest, official art, absurdres, highres)
negative prompt: worst quality, low quality, (censored, bar censor, mosaic censoring), blurry, artistic error, bad anatomy, bad feet, wrong foot, bad hands, bad proportions, bad perspective, bad leg, bad arm, bad neck, bad vulva, bad reflection, bad ass, bad face, english text, chinese text, watermark, multiple views
prompt start: masterpiece, best quality, amazing quality, newest, official art, absurdres, highres
Lora is capable of more than just portraits (quite good at dark and mystical things because of a certain emphasis on the color range)