AI [AI Art] - Show Us Your AI Skill *NO TEENS*

[felldude]

Using Empiric image to guide the XLSD checkpoint and realistic vision as the refiner (I can't load both XL checkpoints)

Spoiler: .45 vs .65 Denoise

You don't have permission to view the spoiler content. Log in or register now.

 

[Jackster]

This AI Stuff is Fun.

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

But man your Disk Space is gone fast with all the Checkpoints and other Stuff you need.

 

[felldude]

This AI Stuff is Fun.
But man your Disk Space is gone fast with all the Checkpoints and other Stuff you need.

Take the number of files needed for your average stable diffusion (Around 35k) if each one of those only took 10 minutes to program it would still take some one just shy of 3 years working 40 hours a week to program them all.

 

[mfcg1754]

I've only just started messing around with SD, but I think I'm starting to get the appeal of all these gacha games.

Spoiler

You don't have permission to view the spoiler content. Log in or register now.

 

[Rumbrandt]

Date at the Tavern

 

[devilkkw]

View attachment 2814182

View attachment 2814183

View attachment 2814185

View attachment 2814186

View attachment 2814187

Wow. i like it, nice result. wait to see other sport like tennis and baseball

 

[felldude]

My new setup generating a 512x512 image with the same SD checkpoint I use as a refiner, up scaling 2x, feeding it in XLSD at .65, then back into the refiner...still around a minute per image and some of the lora comes through.

This is with I think waffles Alison Angel lora.

 

[Microtom]

My new setup generating a 512x512 image with the same SD checkpoint I use as a refiner, up scaling 2x, feeding it in XLSD at .65, then back into the refiner...still around a minute per image and some of the lora comes through.

This is with I think waffles Alison Angel lora.

I find SDXL to be better with composition, but it seems to be the only thing it's better at. Except maybe when the scene concept is pornographic, then it doesn't composite a good image.

So, I usually start with SDXL to create a basic image, then I switch to 1.5 to add details and lewd stuff. Doing an img2img pass on nudity in SDXL also is of no use.

SDXL is almost useless basically. Controlnet XL and intensive pornographic training should make it a lot better tho.

 

[felldude]

I find SDXL to be better with composition, but it seems to be the only thing it's better at. Except maybe when the scene concept is pornographic, then it doesn't composite a good image.

So, I usually start with SDXL to create a basic image, then I switch to 1.5 to add details and lewd stuff. Doing an img2img pass on nudity in SDXL also is of no use.

SDXL is almost useless basically. Controlnet XL and intensive pornographic training should make it a lot better tho.

It takes 3 or 4 seconds to generate and upscale a 512x512 and feed it into SDXL for noise, but I have done both.

You can do much higher native resolution renders in XL, I've done 1600x1280 with no issue...unless you try to feed that back in a SD checkpoint, It goes from 1.5sec/it to on XL to 70sec/per IT on the SD....good ol FP32 and BF32 compared to FP16 and BF16

But if anything explicit is in the image XL is out without training as you said, and I am lacking about 4GB of VRAM to train XL

So it adds about 30 seconds for me to loop in XLSD, one of these images has been run with it and the other run without.

 

[wicked-green-eyes]

 

[FoolsBeGone]

Primary Color check.

Spoiler: black

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: white

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: red

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: blue

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: yellow

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Rainbow

You don't have permission to view the spoiler content. Log in or register now.

The realism (especially in two of them brunette and blonde) is amazing. TBH I think most people seeing these would assume that they were real pictures and not AI drawn. Outstanding!

 

[Microtom]

 

[IncComplex3D]

 

[Synalon]

SDXL can do Nudes I'm just kind of meh about them



This is a very basic minimum prompt picture made with SDXL.

I didn't even bother refining it in img to img with the refiner.

Spoiler: Same Prompts with DreamshaperXL

You don't have permission to view the spoiler content. Log in or register now.

For now I'm still using models based on 1.5, just because controlnet doesn't work yet with SDXL. When it does I'll give it another go.

 

[felldude]

Fantasy Lora for XL SD 1.0 is up for

You must be registered to see the links

 

[sharlotte]

There's a horny beast behind her... (sorry couldn't resist)

 

[felldude]

Cleric visit.
(Using XLSD 1.0 with .7

You must be registered to see the links

and .3

You must be registered to see the links

refinement)

 

[devilkkw]

Summer dress:

Spoiler: Realistic

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Anime

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Hybrid Toon

You don't have permission to view the spoiler content. Log in or register now.

 

[Synalon]

Well I can't get the pose right, and the backgrounds not drawing in. The robes are the wrong colour and trying to add depth put some kind of box around her hand that deforms when I upscale.
I'm sick of working on it now so I'll just post it instead.

 

[Microtom]

Well I can't get the pose right, and the backgrounds not drawing in. The robes are the wrong colour and trying to add depth put some kind of box around her hand that deforms when I upscale.
I'm sick of working on it now so I'll just post it instead.
View attachment 2818344

It looks like you're not using a VAE