Trained from low res screenshots unfortunately. I forgot to include the 1girl tag so if theres any problem with that please comment havent had any problem with that myself.
Tags felinaferal, cat girl, two-tone hair, furry female, body fur, brown fur,
The outfit I think is just uniform but try police uniform and military,
Description
FAQ
Comments (10)
ooh nice!
Thank you so much, she's my fave SW gal.
Some generations come out with partially removed backgrounds, I presume because some of the sources have the subject partially cutout from full pictures and the subject not isolated.
You should probably retrain it without those pictures to make the gens less unpredictable if not prompting for backgrounds/scenery.
a lot of my stuff is edited like that because of other characters getting picked up by the AI. Never had a problem with it before not sure what youre doing for being one of the first people to point it out
Ill see what everyone else thinks and Ill try it too I dont think I had any problems either with backgrounds looking like that. These are low quality screenshots due to it never being in HD
@Aigenerater
Didn't do much of anything. Just generated her with a dress and it came out with a partial background, using base Pony Diffusion 6. Granted, it was just the one gen so far, but I notice you have a partial background in one of your samples. Testing in pony merges it hasn't happend so far.
You should try some background removal tools, or manually isolating the subject from the background completely.
Not sure if it's the same in XL/Pony models, but in my testing doing 1.5 models, so long as you properly describe the traits of secondary subjects (that don't overlap with those of the the main subject of the lora) the ai can still reasonably train the main subject in populated pictures since it tends to focus on consistent visual elements across a dataset (which will tend to be the character you're training if the dataset has enough pictures).
It's not ideal, tho. It requries a lot of manual set up (auto-tagging helps with the initial dataset, but then you have to go over all the pictures to make sure no subject goes undescribed).
@mrseyker try using autism mix. Ive had problems with it with some models that are human. I use to cut images out in photoshop but it became very time consuming and did it this way and somehow never gave me problems.
@Aigenerater After some further testing, I've not seen another instance of the partially removed background yet. It's weird that the very first gen I made had such an obvious instance. Perhaps this is less of an issue than that first impression led me to believe.
Do still suggest doing proper background removals to prevent this odd occurence in the future.
@Aigenerater I mean, you can clearly see it happening here https://civitai.com/images/19892626 It's not a very common issue it would seem, but it does happen, so you should try to clean up your datasets.
An excellent model, as is all your models. Felina is my favorite Swat Kats girl by far, so thank you very much for making this :)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.