Comments

[hidden]

sulph said:

No worries thought it was a little funny. Do you use stable diffusion locally?

No, i was using the wrong tag I think, i have used Yodao fronted. I have edited my post for better tagging but I don't know how to edit AI Metadata. So they are Animagine XL v3.1, DPM2 Karras, 30 Steps, 20 Scale

[hidden]

sulph said:

What the fuck is this, the model is SD1.5 yet you use ponydiffusion scores and SDXL loras, CFG scale: 15 and no vae?

and you know what it came out just fine there no weird purple bloches, the girl looks normal whats the issue here

[hidden]

Kuriomato said:

and you know what it came out just fine there no weird purple bloches, the girl looks normal whats the issue here

Sorry for my angry attitude, i just want to clarify some things:

The model you're using is a Stable Diffusion 1.5 model, and the loras you're using are only meant to work with Stable Diffusion XL, which is another base model that is completely different from 1.5, and they are not being activated since they dont work with sd1.5 so you wouldnt see any difference without them, "score_9, score_8_up, score_7_up" are quality tags used with Pony diffusion (which is a Stable diffusion XL model), in SD1.5 you would use tags like "masterpiece, best quality" which works very well, as for Cfg scale, you might want to lower it down to around 6.5 to 8. I also see that you either dont use a VAE at all or a bad one, try out using kl-f8 anime vae found here: https://civitai.com/models/23906/kl-f8-anime2-vae or one made by StabilityAI: https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/vae-ft-mse-840000-ema-pruned.ckpt these two are put inside this folder: stable-diffusion-webui\models\VAE

Again, sorry for being rude!

[hidden]

Nur said:

Sorry I just dont know, I have not feeling well since a long time ago. I dont want to cause problems to anyone.

If you need help, people are always there for you, talk to your family, friends, a therapist/doctor, hope you feel better soon mate

[hidden]

Heyo, trying to understand your Neg Prompt meta data, looks like you copied in a bunch of extra stuff, but it helps me understand your awesome quality a bit. The "Template" part has all this

`__hairColor__, __hairStyle__, {charcoal black|white|vibrant {red|orange|yellow|green|blue}|pastel {pink|blue|green|yellow|beige}|dark {brown|red|purple|gray}}`

which may be what the system interprets `multicolored hair, curly hair` as, but just curious on that and it looks like you might have used ADetailer during inpainting. Would love to chat about your work flow sometime.

[hidden]

@Sponch
damn, looks like I messed up when copying stuff around, fixed. Usually I try to add the full metadata as a jpeg comment but I'm in the process of switching to comfyui and it's a bit of a mess right now, so I don't remember which posts have what inside.

the stuff in double underscores is wildcards and as you have guessed it picks one of the options for the prompt. I use adetailer pretty much for every image to fix the face. You can find me on aibooru's discord if you want to chat. Or just dm me.

[hidden]

antlers_anon said:

@redd @sulph I love you but c'mon people, you're making me sound like some kind of weird fetish poster. Portable, self-lubricating fleshlights aren't even that uncommon of a kink... I think they aren't at least

Well... I guess with you, that makes 2 people.

1 74 75 76 77 78 79 80 81 82