Indroduction:
Luna is a part of the Adventurer Guild Staff, she is the go-to woman for those visiting the guild. She is a supporting character in the Konosuba series.
There is more Konosuba character coming in future, i hope i'll do all characters i wanted to. But i have almost no experience in character creation.
Recommendations:
1.0 weight
luna \\(konosuba\\)as a support tag
Suggested regular prompt:
luna \\(konosuba\\), blonde hair, folded ponytail, yellow eyesfor appearanceoff shoulder, long sleeves, white shirt, guildmaster shirt, shorts, white sleeves, cleavage, detached collar, red bowfor default outfit
If you want to change clothes on her, lower LoRA's weight a little, 0.7-0.8, there was almost no art of her in different clothes, so, some attributes from her original clothes may appear on other clothes as well.
Credits:
Trained on NAI Anime Full. Dataset of 100 pictures. 30 Epochs. Clip Skip - 2.
Really BIG thanks to:
Some good folks from Discord server
You, for reading, downloading and reviewing this LoRA
Description
Initial release
FAQ
Comments (7)
The \\ in the trigger is bad (ｰｰ') To use it with Automatic, and probably anything else that parses prompts, you have to use
luna \\\\\(konosuba\\\\\)for the trigger (one slash before each slash, and one before each bracket). Just having 'luna' in the prompt works as a weak trigger, but because I'm crazy I did an XYZ grid with AnyLora comparing just 'luna', 4 slashes and 10 slashes, and 10 slashes produced the best results (especially her eyes).
Please use a normal trigger next time f^_^')
| You have to use luna \\\\\(konosuba\\\\\) | - Didn't get any better results with this. But if it so, need an info source. For the first time I hear. As i know, slashes are important to prevent empasis on tag, but nothing more.
*Checked, to be sure. There is a lot of LoRAs with 2 or 4 slashes in training tag. Can't see any problems with them.
| Just having "luna" in the prompt works as a weak trigger | - Of course, in was trained with different tag. And it need other character features to be tagged as well, because it was trained with them.
Accidentally trained LoRA with slashes and fandom, forgot to replace the tag, but never seen any differences, to be honest.
@TroubleDarkness You can read Automatic's code which has a comment that describes the syntax here, https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/5ef669de080814067961f28357256e8fe27544f4/modules/prompt_parser.py#L371
Hi there
I think You mistakenly copied the description of your Lalatina/Darkness Lora.
Hi, actually, LoRA for Luna was first LoRA from Konosuba i've released. I have no idea why there is description from Darkness LoRA. But i remember CivitAI was lagging then. Thanks for pointing that out, will fix it now.
UPD: Fixed.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.