Young Zelda from Ocarina of Time is now remade for new models! The SD1.5 version makes great gens, so I was surprised no one else had posted a pony version for quite a while. Same goes for Illustrious.
NEW v3 (Illustrious)!
Ports OoT Zelda to a newer and more dynamic model. It's trained on Illustrious, but works far better in NTRmix from my experience. Notably, the triforce and hat may not come out as consistently as it does on Pony models, and it seems to depend greatly on presence or absence of certain clothing tags. For example, trying to generate Zelda with a t-shirt or jersey may force the hat into a baseball cap with a brim. To combat this, try upping the strength of the LoRA or main ootyoungzelda tag, or add offsetting tags like (headband:0.5) to remove the cap before it starts setting in.
v2 (Pony):
Better control in clothing, strength tuned down to mesh with other LoRAs, and little impact on gen quality otherwise. See version notes for details.
v1 (Pony):
This LoRA does come with some quirks. It was my first time training a LoRA, so the parameters are probably sub-optimal. I used a dataset of 58 images comprising a selection of official art, fan art, and hi-res screenshots from both versions of the game. These images were then cleaned up, cropped and tagged, then ran through the trainer using base Pony and mostly stock parameters.
The output is a little overtuned on certain details, like clothing and presence of Zelda's hat and extra triforces, but is still usable in most scenarios. I recommend playing with the LoRA strength (between 0.6 and 0.9) and the strength of the base "ootyoungzelda" tag (1.0-1.2), erring on the low side for both. You may need to juice other appearance tags a bit more than usual to get them to show. Notably, the sleeves and shoulders will almost certainly mimic the official look without some hefty weighting otherwise.
Also note that this LoRA will typically generate a roughly teenage Zelda without any additional influence. For her canonical younger appearance, you may need additional tags or LoRAs.
Description
Version 2 is more controllable, and is less insistent on certain details such as the necklace/brooch and the hat. By default, Zelda's hat may or may not show up, so add "hat" to the positives or negatives. The same is true for "long sleeves" and "necklace"/"brooch". Strength/tag weights may need to be a little higher to match v1 behavior, which should put it in line with most other LoRAs since v1 was overtuned. Some details from BotW Zelda may show through at lower weights, such as a bob cut forming instead of the hat, or the presence of hair clips or hair hanging down in front of the ears. Such details are not in the LoRA training data, so check your base model and weight accordingly. Finally, triforces are not quite as clean in this version. v1 was very good about the emblem in the hat and low on the tabard, but v2 tends to artifact a tiny bit more. It'll take your triforce success rate from 90% down to about 80%.
FAQ
Details
Files
Available On (6 platforms)
Same model published on other platforms. May have additional downloads or version variants.