BEST GIRL is here! 🎉
Marnie is my favorite Pokegirl.
Please post pictures of Marnie🙇
Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
Please checkout all my Pokemon Loras: https://civarchive.com/collections/261
Description
zzMarnie, aqua eyes, black hair, medium hair, asymmetrical bangs, twintails,
black choker, red ribbon, pink dress, jewelry, black jacket, open clothes, long sleeves,
,<lora:MarniePokemon_NAI_v2:1.0>,
FAQ
Comments (9)
I'm in love with her eyes! Help!!
There is no escape. Once you fall for Marnie its permanent love! LOL
Your posts look amazing! Spot coloring (not sure if this is the right term) just her eyes gives a beautiful effect.
@CitronLegacy Yeah, I can see that now. 🤣 The whole point of the "spot coloring" was so that everyone focuses on the fact that those are the greatest eyes ever! Stare into them and lose yourself.
Was the training data identical to Illustrious for ZIT? Curious how well it does tag based datasets.
Thanks for the comment on my favorite Pokemon Lora!
I made a new dataset for ZIT. Its way smaller than my Illustrious dataset. Here are is a preview of all the images I used: https://x.com/Citron_Legacy/status/1997757877206098122/photo/2
As for using tags instead of captions it seems good. I'm tentatively thinking using tags is fine but I want to make a lot more tag based loras before I make any conclusions.
I have another 5 character Loras trained on tags that I need to post so we'll have some more data soon.
@CitronLegacy Thanks. I guess the ultimate way to check is to train one pic set twice, once with tags and once with captions, and compare the outcome. I may do that, though one style with a smaller than desirable dataset size is the only natural language dataset I have (excluding the joke Ultra Lord one), and it may ultimately depend on my (terrible) natural language writing skills.
@NanashiAnon I'm horrible at natural language writing so I have to rely on CivitAI's auto-tagger.
Once the CivitAI trainer stops breaking I'll be able to do more tests. I had 15 more trainings fail so its like 40% fail rate right now. Luckily the buzz does get refunded its just a lot of waiting.
I think if tagged datasets seems to preform less well, then the next test would be to increase the image repeats or Network Dim.
I trained this Misty lora with CivitAI's auto-captions (I didn't even look at the results I just used it lol) and I am able to prompt it with tags so thats at least a good fall back https://civitai.com/models/131452
MARNIE MARNIE MARNIE MARNIE 😍😍😍😍😍😍😍😍😍😍😍
YES! I've been early waiting to post this!
ALSO she reached 1k likes! I'm so proud of my girl!
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.