Sign In

Princess Luna (humanized)

232

Download

1 variant available

SafeTensor

72.13 MB

Verified:

Type
LoRA
Stats

203

232

23

Reviews
Published

Apr 16, 2024

Base Model

SD 1.5

Training
Epochs: 8
Usage Tips
Clip Skip: 2
Strength: 0.9
Hash
AutoV2
F9978C6E6E
Trigger Words
princess_luna
black crown
single horn
sparkling hair
regalia
default creator card background decoration
Followers - 59

59

Likes - 410

410

00002-478519546-princess_luna wearing casual clothes, holding a game controller, clenched teeth, toon style _lora_luna-08_0.9_, masterpiece, bes.png

Princess Luna is a character from My Little Pony: Friendship is Magic. This her humanized version.

The main trigger is princess_luna, with optional black crown, single horn, regalia, and sparkling hair.

Weight 0.9 is recommended. You might need to increase or decrease the individual weights of the activation tags to get better results.

Don't overdo with negative embeddings, they might remove color from Luna's multicolored hair.

Activation tags

1) princess_luna is the main trigger. Other activation tags can sometimes bleed into the image, so consider adding them into the negative if you don't want them.

2) black crown has been fine-tuned to look like the crown Luna wears in the show.

3) single horn gives Luna her horn.

4) regalia is the necklace/choker jewelry artists tend to draw to imitate the regal harness Luna wears in the show.

5) sparkling hair to create stars in Luna's hair. This is such a common element that you'll usually get it anyway. Probably more useful in the negatives if you don't want it.

Extra tags

1) black crescent crown to create a variant with the moon in the middle, but only if you're lucky.

2) cutie mark to attempt adding Luna's cutie mark. Won't always work or look good, and might end up being generated on clothes.

3) crescent earrings and crescent necklace to add matching jewelry for the princess of the night.

4) blue skin and colored skin together to generated the show-accurate-ish skin tone.

Misc.

Luckily this Lora wasn't that hard to create, so I went a bit further with it than with Princess Celestia.

Trained on 173 images (x6 repeats), with a mix of person and anthro used as a class. 8 epochs, 64 DIM / 32 Alpha. UNET 2e-4, TEXT 4e-5, Cosine with restarts 3. Base model used for training was animefull_final, but the example images were created on AbyssOrangeMix3.

My fourth attempt at making a LoRA. Feedback is welcome.