Firis Mislud character LoRA
Training based on the official art book and in-game 3D model.
Having very good facial and hair generation results.
Firis clothes are more complex than other Atelier characters. So, when you generate your images, clothes may have missing or wrong detail.
The cause of this problem is mainly due to the Stable Diffusion model used to generate the images and my training set.
Firis's original official concept artworks, due to the artist's style, there are errors and missing details in many of them, which can be easily learned by the machine. Moreover, the colors and lines in the original paintings are often not particularly obvious.
P.S. I plan to redraw some training set images manually to improve training results.
Description
Slightly Revised version based on last version
Recommended weight: 0.8, DO NOT use so low weight
Recommended CFG: 5.5-6.5
Adjustment:
better neckbone and cloth rope
improved clothes
better hair ornament (using prompt "crystal" "circlet" may help generate better image)
Still existing issue:
in some Stable Diffusion model or certain VAE, you may often generate full black image
if you are trying to use this LoRA in low resolution (lower than 512*512), you may often notice missing details
still having problems with full body image generation, Non-character details often appear on the results.
FAQ
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.