probably need to add/change your VAE

试试插件adetailer，修脸神器，内置的面部修复不要用了

Try putting stuff like "loli, child" in the negative prompt and "adult" in the positive prompts. But yeah you should also check if it's not your lora creating images of children

@Meina @CivitO_O i tried without any loras and had old in positive prompt and child, loli in negative but it still happens :/

What kind of prompts are you using? I've found that, for certain models, certain trigger words tend to be correlated with the age. For instance, triggers like "small breasts" or "huge breasts" can sometimes have an effect on the apparent age of the subject.

@mangaka92 ah that would make sense, i use small breasts and skinny usually

@sneksnek24 Yup, that would probably do it.

@sneksnek24 Having "small breasts" will always end up generating REALLY young girls, avoid using that if you don't like it as the default "cup size" is already pretty small to my taste... lol

@Meina Sorry, still a bit of a novice. So, does that mean I should not use a VAE with the Meina models since the VAE is already part of the model?

@psmallwood217549 I tested with and without VAE. The result is the same, which means that when using this model, its use is not required

@Meina Thank youuuuuuu, I finally figured it out, love your model, great all around!!!

Yes, there is! Hires. fix is essentially just img2img, with 2 differences: 1. You can choose your upscaler, 2. The steps are fixed and not multiplied by the denoising strength like img2img does. Both of these can also be done in img2img. In Settings > img2img you will find "Upscaler for img2img" and "With img2img, do exactly the amount of steps the slider specifies". Activating the fixed steps and choosing the same upscaler as hires fix will make img2img behave exactly the same as hires fix. If you want to get the exact same image from img2img that hires fix does, just use the same seed in img2img that you used in txt2img and use Euler A, as that's what hires fix does

@Meina when I use that VAE I get very saturated images even if I'm not using any Lora, may I ask if you know the reason for that? Thank you in advance

Sadly, a lot of models are trained exclusively on attractive asian females, meaning they struggle to generate anything else.

I agree with the comment above me but I can manage to make a femboy just fine, probably need to add 1girl and other girly features into the negative prompt if you want the MAN part

I think there's a extension for that somewhere but I forgot the name sorry

Generally, you want to specify the style of image, pose, focal subject, details about the focal subject, setting, background details including background characters, in roughly that order. Then put parenthesis around the nouns and adjectives that pertain to the focal subject.

e.g. "shoujo anime splash art viewed from above of a (seated woman), with (sparkling eyes, lipstick, huge boobs), wearing a (white sundress), (holding a teacup), (sitting) at a table in a rose garden, hedge mazes and butlers in the background"

You can get ideas for subjects and details you like by reading the prompts from other people's images that you like, until you get the hang of it.

If you are just getting started, the best thing to do is to browse the pictures generated by others here, and if you like one, check out the prompt they used by clicking on the circled "i" in the lower right corner.

u can use last

@tkachevdanila6899 TY

If you know what baked vae means, it means it's already in and it's already answered way back which is using automatic

"None" parameter will use baked-in VAE if there it is.
"Automatic" parameter will use VAE with the same name as model's name.

Even if VAE is baked in model, you can use any VAE you want.

@Noxy_  Thanks

@TroubleDarkness Thanks 

me too

too >:(

I experimented with a lot of things.

Apparently some Lora still does not work well.

This right here, at least tips or things to do rather than leaving us hanging

@Noxy_ I'm late to this, but have you guys tried using Adetailer to fix hands?

@Maxx_ It doesn't fix most of the results and sometimes it doesn't work at ALL, so it's a random dice roll now

Although, using that doesn't work for me anymore

@Noxy_  It is always a roll of the dice, but typically, (bad_hand) (deformed_hand) (ugly_hand) can help, sometimes just one, sometimes all three, sometimes with one or more of these getting 1.2-1.5 weight, it varies by checkpoint and seed.

Yes, doesn't matter. better if use xl models. Check my profile