Princess Zelda from the 1989 cartoon series "The Legend of Zelda" (originally shown as part of the Super Mario Brothers Super Show in North America). Also included are outfits from her later appearances during "Captain N: The Game Master."

UPDATE - 25 March 2025 - Added Illustrious version

Suggested strength: 0.7 - 0.9

To create this LoRA, I set out on my most adventuresome quest yet. I wanted to see if I could create a character LoRA with a huge dataset and still get a good result, so this LoRA has a 536 image dataset trained to 3,431 steps. I think the outcome was good, but there's still opportunity for improvement. I couldn't locate any HD sources for the shows, so I had to upscale low res images from a 35-year old hand drawn cartoon and then clean them up by hand in Krita. As a result, a lot of the training images are lower quality than I'd prefer. Now that I have the LoRA to help me, I can probably do a better job repairing them for version 2.

I included all six screen worn outfits from the 1989 series, including my favorite: the evil mirror clone from the episode Doppelganger. The evil clone basically wears Zelda's normal outfit but with a darker color scheme as she tries to seduce Link. I also included two outfits from the Captain N series. The art style is a little different and she's meant to be a few years older and a little more mature, so her face and hairstyle will differ from the other generations if you prompt the Captain N outfit or the Ballgown.

I also tried to create tags for whole outfits in addition to tagging individual garments, but the whole outfit tags seemed to degrade the output quality during testing. Maybe they're overcooking the prompt during generation? I recommend just using the main trigger and individual garment tags instead. I'll list the "whole outfit triggers" at the end of the article below if you want to play with them but really I recommend just prompting "z3ld4" and individual garments instead.

ALL OF TREASURES

• Main trigger: z3ld4

• General Triggers: blonde hair, long hair, makeup, red lips, eyeshadow, pointy ears

• Outfits:

  1. Standard Outfit: blue tiara, purple blouse, periwinkle vest, pink pants, brown thighhighs, gold arm cuffs, belt pouch, red earrings, brown belt

  2. Doppelganger (Evil Mirror Clone) Outfit: blue tiara, black blouse, red vest, black pants, brown thighhighs, gold arm cuffs, belt pouch, red earrings, brown belt

  3. Pink Nightgown: pink nightgown, shoulderless, rose garland, barefoot

  4. Magenta Bathrobe (two headwear options): magenta bathrobe, red earrings, barefoot, PICK ONE: rose garland -or- blue tiara

  5. Orange Gown: orange gown, shoulderless, orange slippers, blue tiara, red earrings

  6. Pink Swimsuit: pink swimsuit, one-piece swimsuit, cutout cleavage, blue tiara, red earrings, barefoot

  7. Captain N Attire: purple crop top, blue diadem, sidelocks, vest, midriff, purple glove, gold bracer, black belt, pink pants, pelvic curtain, brown greaves, blue shoes, gold kneepads, gold elbow pad, belt pouch

  8. Ballgown from Captain N: blue ballgown, white cape, blue diadem, sidelocks, gold necklace, gold earrings, sash, purple sleeves

As usual with my LoRAs, everything is well-trained enough that you should be able to mix-and-match at will. So if you want to wear the rose garland with her standard outfit or the blue tiara while nude, you should be able to prompt for things like that. You should also be able to have her wear random clothes with no problems.

As promised, here are the experimental "whole outfit" triggers that I trained into the LoRA. As I mentioned above, my testing results with these were poor. Feel free to experiment with them if you wish, but I'll probably remove them entirely from version 2:

• z3ld4standard

• z3ld4pink

• z3ld4black

• z3ld4magenta

• z3ld4orange

• z3ld4swim

• z3ld4capn

• z3ld4ball