If you want just normal Gardevoir do the following and make changes as needed. See note at the bottom if you want a very pokemon style
Default Gardevoir:
gardevoir, white dress, creatures \(company\), game freak, nintendo, pokemon, pokemon \(game\), bangs, colored skin, female focus, flat chest, gen 3 pokemon, green hair, green skin, hair over one eye, multicolored skin, pokemon \(creature\), red eyes, short hair, two-tone skin, white skin
hair styles (Do not use with the Default Gardevoir Prompt just the 'gardevoir' tag):
side ponytail
ponytail
twintails
long hair
hair bun
any hair color
artists (Use at the end of the Default Gardevoir Prompt):
(kionant:0.8)
(drunkoak:0.8)/(drunkoak:0.9)
fir3born (couldn't decide a best weight), flat chest, nsfw
concepts (Use at the end of the Default Gardevoir Prompt unless stated otherwise):
mega gardevoir
(dr.voir:1.2), blue hair, glasses, labcoat, black panties (don't use default just 'gardevoir' alternatively see below)
clothes of any kind (maid works well)
pokemon unite outfit (see below for prompt)
mecha
swap 1girl to 1boy and add penis to get male gardevoir (don't use this with the default prompt)
Shiny Mega Gardevoir:
gardevoir, creatures \(company\), game freak, nintendo, pokemon, pokemon \(game\), bangs, colored skin, female focus, gen 3 pokemon, blue hair, blue skin, hair over one eye, multicolored skin, pokemon \(creature\), orange eyes, short hair, two-tone skin, white skin, black dress, mega gardevoir
Full Dr.Voir Prompt:
dr.voir, gardevoir, creatures (company), game freak, nintendo, pokemon, alternate color, bangs, blue hair, blue skin, breasts, choker, collar, collared shirt, colored skin, glasses, hair between eyes, hair ornament, hairclip, labcoat, large breasts, mole, mole under eye, multicolored skin, open clothes, open shirt, pokemon \(creature\), red-framed eyewear, red eyes, semi-rimless eyewear, shiny pokemon, shiny skin, shirt, short hair, short sleeves, tareme, two-tone skin, white shirt, white skin
Pokemon Unite Outfit:
gardevoir, gardevoir \(fashionable\), creatures \(company\), game freak, nintendo, pokemon, pokemon \(game\), pokemon unite, bangs, bob cut, bonnet, capelet, clothed pokemon, colored skin, female focus, flat chest, flower, gen 3 pokemon, green hair, green skin, hair over one eye, hat, hat flower, multicolored skin, official alternate costume, one eye covered, pink trim, pokemon \(creature\), red eyes, red flower, short hair, two-tone skin, white capelet, white headwear, white skin
As always:
Trained on Anything v3
Demos are Anything v4.5 with 0.8 weight LoRA
Note:
I ran out of room for included images but using the default prompt along with the Harada Takehito LoRA I released (and putting harada takehito at the end of the prompt) does a surprisingly good job of emulating the correct style about 80% of the time.
Description
Much larger dataset that allows for a lot of control
FAQ
Comments (34)
I am going to do terrible things with this Lora. Just terrible things. And then the second I get my hands on a sonic lora the disgusting things I shall do with it shall shock this world
Ayo. Wtf. im gussing youll be using the vore and feet loras for sonic. Lmao. jk
wife generator
Hello! Quick question. Do you (or anyone else whose been using the model) know a consistent way to hide or reduce the red horn things on her chest? Usually just specifying clothing is enough but sometimes its stubborn, and I was wondering if there's a quick negative prompt I could use to minimize it.
I'm not sure to be honest. You could try using 'red' in the negatives maybe. I was trying to stay close-ish to the model for creation. Maybe in the future I can retrain including images that don't have it and tag them specifically. Sorry I can't be of much help
I was about to point you toward some of the generations in the gallery, but then I realized you were the same person who made them haha. I should really pay more attention to usernames
@thesauceer Lol you're fine! :P And yeah its no big deal, I ended up turning down the model strength. Its not a perfect solution but it gets the job done. Thank you!
I seem to be absolutely unable to get great results with this, far from as good as the example images, I don't know what I'm doing wrong or if it's some problem with a vae or something, but if anyone could help me that would be great.
What is the current set up your using? Have you tried to recreate one of the images using the exact set of inputs and models?
I have indeed tried to recreate it, is it perhaps something to do with lora strength?
@thesauceer I at the very least tried to perfectly replicate the example images but it didn't come out well, I just tried again and it went a little better and I am more fine with it now but I think the main issue is lora strength, I'll keep searching for other potential causes of the issues I had before
Yeah I used the LoRA at 0.8. Also, just to be sure, you have the easy negative embed right? The sample images were made using it, but if you don't have the actually embed and are just using the text in the negatives it won't work the same. Also, all of these images are done with a latent upscale of 1.5. The original dimensions are 512x768. I started typing this before checking what info is or isn't available so I'm sorry if there is some repeated info!
@thesauceer alright, thanks I wasn't sure about the lora strength so I usually had it at about 0.6-0.9, thanks for telling me, and I'm not sure if it would've affected it but I was upscaling by 2 from 384x512, yet again ty for the info, I'll try it again like this!
@bananavee0924 Ah yes the different starting size and upscale amount would change a lot. Hopefully you got it working. If not the let me know!
@thesauceer it's working much better now, thanks so much! And sorry for bothering you for help lol
@bananavee0924 Glad you got it! Enjoy!
Any plans on Salazzle?
There weren't, but every comment I receive basically acts as a list of things I look to when I want to make a LoRA, so it's a possibility now haha
Hello, what model do you use for the basis of image generation?
The examples provided all used Anything v4.5
What black magic is used here to make it work so well on realistic models? I thought I'd try this lora for shits and giggles while I was using Henmix and got a weirdly disproportionate amount of actually good results. That never happens with other kinds of cartoony lora's.
Oh I've never tried it on a realistic one. I'll need to give that a shot at some point. The only thing I know that was particularly unique about this LoRA was the number of training images due to her popularity. The dataset was 957 images. The only things I've trained with more were a couple of style LoRAs that I'm not sure I've released here or not. Anyway, glad to hear that it's more versatile than expected!
@thesauceer With more time and experience I realize that it's also partially Henmix being pretty receptive to cartoony/anime loras, but yeah this Lora still worked particularly well. I guess it does make sense considering the sheer quantity of Gardevoir stuff out there. Good job!
@Speedhoven Ah that would make sense! Either way I'm glad you find it useful!
What the F Did You Just Bring Upon This Cursed Land
Enjoy! Haha
What checkpoint should i use to get similar results?
I used Anything v4.5 for the example images
It is truly amazing that, after 8 months, I still can't find a Gardevoir model better than yours.
I'm glad it still holds up!
How can I use this model? I tried to use it but due to my settings. I can't use her.
Nice Model you made
Yo, this model is awesome, any plans to make an illustrious version??
I'd love to, but it was made so early on into the image gen timeline that I didn't think to keep the training data. If I was to remake one now I'm not sure how it would turn out. It was made before auto tagging so all the tags were added manually and the photos I used were edited to better suit the end result. Due to all of that I'm not sure I could make lightning strike twice sadly.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.