Hi there, today i am gonna share with you another character lora. This time the character is Lunaria, the heroine of "shinmai megami no oshigoto wa tsurai" (it's a relatively recent doujinshi - more precisely a doujin of last year).<h3 id="heading-4034">Let's talk about the lora</h3>It works good between 0.6 and 1.I suggest a weight of 0.7 for most cases, 0.6 when you need to make the lora even more compatible with other loras, and higher values when you want to produce an "illustration-like effect" on your image.The tags you might wanna use are:<pre><code>Lunaria &lt;-- the activation tag
Blonde hair, short hair, blue eyes &lt;-- for the classic character palette
Tiara, leotard &lt;-- for the "classic outfit"

ps. the detached collar (the one from the shirt) is part of the classic outfit</code></pre>The are few precisation i have to make.The dataset contained a lot of images with a collar (the leather collar), so it might happen that the collar generation is a bit unstable, this is easily fixable by tweaking the prompt; in any case here are some tags you might wanna put in the negative when generating images that have no collar:<pre><code>detached collar, bra strap, collar
most of times these works as they are, if the seed is particoularly unlucky you might need to place them in brackets "(",")".</code></pre>I should've included a white bow on the right-side of the head, but due to dataset limitation, it wasn't correctly learnt by the model (so the generation of that item is kinda inconsistent).I hope this is of use for someone of you.As always you can support me by sharing your generation, reacting to images, reviewing the model or by donating buzz.

Lunaria - shinmai megami no oshigoto wa tsurai

This is a lora trained on mostly monochrome images, so the best results are produced in manga style (it is also one of the few character lora that support a real "from_back" images).

Files with SHA256: 18d149dd2d10ab05bd30f96ac6c70dd0f02e19819639749ed72e3cf64567dbf5