Sign In

Luna ( Zero Escape , Virtue's Last Reward ) [SDXL]

4
32
0
Verified:
SafeTensor
Type
LoRA
Stats
32
0
Reviews
Published
Dec 9, 2024
Base Model
SDXL 1.0
Training
Steps: 1,575
Epochs: 30
Usage Tips
Clip Skip: 1
Hash
AutoV2
7EC25C47C0
default creator card background decoration
Second Birthday Badge
emerjk
Created on Civitai

Luna from Virtue's Last Reward.

Prompt style

This lora was trained in narrative, rather than tag, style. You can go with the narrative promps using the examples below.

Prompts

Standard outfit: "Luna has droopy blue eyes and thin eyebrows. Luna's coral orange hair is tied in a bun with a braid and with two long strands on each side of her face in the front, accompanied with short, side-swept bangs. Luna wears a light purple, long-sleeved blouse under a white vest and a long, dark purple skirt with a white geometric design near the bottom. The skirt is tied around her waist with a ribbon. She also has purple stockings and white slippers. She wears a large bracelet on her left wrist. Her necklace is a caged bluebird."

Bikini: "A photo of Luna from the Zero Escape series. Luna is wearing a light purple bikini top and matching bottom. Luna's coral orange hair is tied in a bun with a braid and with two long strands on each side of her face in the front, accompanied with short, side-swept bangs. Luna is wearing a bracelet on her left wrist."

Additional prompt options

Hands to forehead: That odd gesture she does when putting her hands up to her head. It didn't come out as well as I'd hoped, presumably due to the lower amount of training data. But it'll sometimes just come out if she's described as having a shy or embarrassed expression. But you can also try to trigger it with these tags, "shy, embarrassed, hands raised to forehead"

Styles

The style options really didn't come out very strongly. However, you can try the following.

VLR sprite style: (Leans more to the in-game graphics): "VLR sprite style."

Anime style from the short Virtue's Last Reward OVA. (Fairly weak due to lack of training data): "VLR anime style."

A more 3D look, trained from cosplay images: "A photograph of Luna."

Issues

The biggest issue I've found is that the lora doesn't replicate that 'hands to forehead' pose Luna does in the game as well as I'd like.

Her birdcage pendant and nonary game bracelet also tend to be inconsistent. It's fairly normal for the training data to have both be blurry and suggest more than show the details. Sadly, that carries over to the lora. The bracelet is sometimes rendered more like a watch, or on both arms.

Models

The lora seems to be a little more highly variable to both model and prompt style then I'm used to. I'd advise playing around with it a little using different models or prompt styles if you're not getting good results. For basic 2D, I'd advise trying AAM XL. AniVerse XL can produce a nice semi-photorealistic effect that's about midway between 3D model and live-action.

Additional notes

The major challenge with this lora was the lack of training data. There's not nearly as much or as wide a variety in Luna art out there as I'd like. I tried to keep an even balance between in-game art, official illustrations, fan art, and peppered in some cosplay images. I think it worked out OK, but just keep that in mind.

The bikini outfit option is in there because I found what seems to be officially branded artwork of her wearing it. Though I don't know where it's from. But like I said, lack of training data is an issue so if it's official and higher quality then it goes in.