I debated whether or not to handle the "real" Twili with this Lora.
Eventually I decided to focus on making something that could turn existing characters in to Twili.

Some day I might make another Lora which is Twili from the game.

Hi Magof :)
I remember you asking about this before. I am working on a guide to make a Lora that is good every time. Sometimes my OC Loras are good and sometimes they aren't. (The one I was making last time we talked didn't turn out well) I'll publish a guide once its reliable.

Until then, what kind of OC are you trying to prompt?

Here is really long description of the process to generate OCs

I just posted at least 30 images of Maxetto's OC using the prompt for the OC. (You can find links to most my posts here: https://civitai.com/bounties/660 )

Lets talk about the images in this post: https://civitai.com/posts/1045334
As you can see in the posts certain aspects of the OC are consistent simply by using the same prompt

1girl, blue eyes, smile, very long hair, blonde hair, long blonde hair,( french braid:1.1), bangs, medium breasts, hair ribbon, frilled choker, criss-cross halter, (blue dress:1.1), sleeveless dress, high-waist skirt, backless dress, waist bow, black detached sleeves, frilled sleeves,

She always has the same long blonde hair, eyes, and french braid. She usually has detached sleeves and a criss-cross halter but sometimes she doesnt. She almost never has the same dress.

So the problem here is that the AI doesn't understand the specifics of her outfit, however if you generate enough images you get a few that are consistent. This image post has 20 images but I generated at least 40.

Its possible to do this with out a trained model but it will take a ton of image generations.

Therefore we know making a Lora (or any model) is best, but the problem is these images are very different so we can't use them for training data.

What we would want to do is take a few of these images that have characteristics we link and edit them to be perfect. For example, the detached sleeves are inconsistent so I'd Photoshop (technically I edit images with Paint Tool SAI not Photoshop) and inpaint them to get them perfect. I like the white puffy sleeves so I'll not edit those but I will edit the others until they have the puffy sleeves.

I'd do this process until I had at least 10 images with consistent outfits to train a Lora.