V8
This is the 8th attempt at a better Jessie lora than my previous one. No point uploading crappy in-between loras.
It works pretty good overall but it can get stuck or shifts under certain circumstances.
The team rocket shirt seems to be baked in pretty hard unless tagged out with different clothes.
Portrait definitely does better than landscape. If landscape keeps reverting your images to the fuzzy DVD look. Use these negative tags:
"1990s \(style\), anime screencap, official style,"
NSFW seems to work fine, but it can sometimes shift away from the classic visual appearance. The eye shape is the most important feature of a genuine jessie pic!
Main trigger tag to use is:
"jessiepkl"
A very heavy trigger word to use if the character is shifting:
"Team rocket uniform"
To get the classic look of the season 1 DVD:
"official style, anime screencap"
Use it how you want, it is but another drop of Jessie in what seems like an ever growing ocean of Loras.
My previews were made with:
anima-base-v1.0
CFG: 4
Steps: 30
Sampler: Euler Ancestral
Scheduler: beta57
Width/ Height: 1256 : 840
Positive prompt:
"masterpiece, best quality, score_7, jessiepk, 1girl, ............................"
Negative Prompt:
"worst quality, low quality, score_1, score_2, score_3, artist name, blurry, censored, signature, monochrome,"
V1
This is my first Lora trained on Jessie's image from Pokemon Season 1. It definitely influences the style of the output.
It gives your images a nice 'original' Pokemon DVD look. It seems reasonably flexible at times, but once you get lower to the ground it seems to get a bit more rigid. It Seems to work with NSFW too.
Do expect a bit of jank and stiffness, I didn't feel too confident with what I was doing. I think the next one will be better.
Try it out if you like. Trigger word should be 'jessiepkl'. But other tags that are useful are: team rocket, team rocket uniform, black gloves, black thighhighs, heels
I like to use:
Steps: 50
CFG: 5
Sampler: Euler Ancestral
Scheduler: beta57
If you think you can do better or want to try it out, I have attached the dataset of screencaps. Create more of Jessie! :)
Description
Reduced dataset, captioned along with tags. Can generate clean




