/loli/ - AI Generated Lolis

Name
Options
Subject
Message	Max message length: 12000
files	Drag files here to upload or click here to select them 0.00 / 50.00 MB Max file size: 32.00 MB Total max file size: 50.00 MB Max files: 5 Supported file types: GIF, JPG, PNG, WebM, OGG, and more

E-mail
Password	(used to delete files and posts)
Misc

Anonymous 09/29/2022 (Thu) 13:28:15 No. 4001

Now share your lolis and your tips for better lolis. The power is yours!

Anonymous 09/29/2022 (Thu) 22:01:58 No. 4002

Fucking glorious.

Anonymous 09/29/2022 (Thu) 23:22:19 No. 4004

Stable Diffusion actually recognizes art styles by the name of the artist. There are a bunch here: https://rentry.org/artists_sd-v1-4 I managed to make these muscle lolis by adding Boris Vallejo to the prompt.

Anonymous 09/30/2022 (Fri) 03:25:51 No. 4005

From /v/.

Anonymous 09/30/2022 (Fri) 14:38:06 No. 4006

You can blend different art styles together by using inpaint with a denoising strength of around 0.3 Basically you generate a pic in one art style, then you inpaint the part you want to change and change the prompt to a different art style. Since the denoising strength is so low, the AI will try to blend the art styles together and it can give great results. Pic 1 is Albert Lynch mixed with Ufotable. Pic 2 is Tsuneo Sanda mixed with Murata. Pic 3 is Tsunako mixed with H. R. Giger.

Anonymous 09/30/2022 (Fri) 20:27:40 No. 4007

Messing around with different art styles is a lot of fun.

Anonymous 10/01/2022 (Sat) 17:42:15 No. 4014

Anonymous 10/02/2022 (Sun) 14:41:42 No. 4016

Anonymous 10/02/2022 (Sun) 17:27:52 No. 4017

>>4016 Oh shit that little android girl is great.

Anonymous 10/02/2022 (Sun) 23:24:00 No. 4018

>>4017 What level of CLANG are you at?

Anonymous 10/03/2022 (Mon) 00:53:35 No. 4019

>>4018 2 for me, mainly because she seems to have two extra limbs (or a strange gun) in her back. 3 and 4 would make great soldiers in a loli-themed sci-fi game.

Anonymous 10/03/2022 (Mon) 17:26:15 No. 4020

Setting the txt2img resolution to 448x704 instead of the default 512x512 helps avoid cut-off heads.

Anonymous 10/04/2022 (Tue) 03:21:44 No. 4022

From /v/.

Anonymous 10/04/2022 (Tue) 20:46:04 No. 4023

UPDATED LIST OF FINETUNED MODELS https://rentry.org/sdmodels Includes new versions of the waifu model as well as lewd models, furry models, etc.

Anonymous 10/05/2022 (Wed) 01:08:11 No. 4024

From https://www.pixiv.net/en/users/681190

Anonymous 10/05/2022 (Wed) 01:08:34 No. 4025

Anonymous 10/05/2022 (Wed) 01:09:08 No. 4026

Anonymous 10/05/2022 (Wed) 01:11:54 No. 4027

From https://www.pixiv.net/en/users/86321977

Anonymous 10/05/2022 (Wed) 01:12:31 No. 4028

Anonymous 10/05/2022 (Wed) 01:14:05 No. 4029

Anonymous 10/05/2022 (Wed) 01:15:19 No. 4030

Anonymous 10/05/2022 (Wed) 01:16:47 No. 4031

Anonymous 10/05/2022 (Wed) 11:18:45 No. 4032

Can someone try to make a cartoon Loli (not anime)? I want to know if the AI will be able to create something like that?

Anonymous 10/05/2022 (Wed) 16:02:27 No. 4033

Anonymous 10/05/2022 (Wed) 17:39:19 No. 4035

>>4032 >>>/delicious/28633 fucked up the first reply

Anonymous 10/05/2022 (Wed) 18:36:29 No. 4036

>>4032 Emulating different art styles is actually very easy for the AI, so yeah you can render western cartoon lolis. I'm using the waifu model, which is heavily biased toward anime faces, so these faces are going to look more anime than in the standard model, but you can see that specifying an artist in the prompt really makes a big difference. I used the same prompt with different western artists for all of these: 1-((8-year-old girl)), revealing costume, navel, (((by Barry Windsor-Smith))), 4k, 1girl 2-((8-year-old girl)), revealing costume, navel, (((by Ethan Van Sciver))), 4k, 1girl 3-((8-year-old girl)), revealing costume, navel, (((by Frank Frazetta))), 4k, 1girl 4-((8-year-old girl)), revealing costume, navel, (((by Don Bluth))), 4k, 1girl 5-((8-year-old girl)), revealing costume, navel, (((by Jay Anacleto))), 4k, 1girl And yes, funnily enough, the (((jewish echo))) in the prompt does make a big difference, the AI will actually prioritize whatever part of the prompt has the most parentheses around it.

Anonymous 10/05/2022 (Wed) 19:30:51 No. 4038

So I tried that furry module and uhh... this happened.

Anonymous 10/06/2022 (Thu) 00:13:44 No. 4040

>>4025 >>4026 >Censor bars What the fuck?

Anonymous 10/06/2022 (Thu) 00:52:33 No. 4041

>>4028 >4th pic Jesus fucking Christ. I want to rub my cock on her fluffy hair. And are those bloomers she's wearing? I think I've finally found a reason to say I'm glad to be alive during this era. This is fucking huge. This is only the beginning, and it's already at a level where it will have an effect on the world of art in similar magnitude and nature to that of the invention of photography. These lolis are world class history being made before our eyes.

Anonymous 10/06/2022 (Thu) 01:14:14 No. 4042

>>4040 Maybe he's a nip and has to follow their dumb censorship laws. The AI doesn't generate censor bars if that's what you were thinking.

Anonymous 10/06/2022 (Thu) 23:15:34 No. 4043

I'm getting better results with the standard model than with the waifu model. The waifu model tends to make all the shading flat and tries to force anime eyes into every art style, giving it a creepy eye-within-eye effect, but with the standard model I'm finding it much easier to blend different art styles with img2img and inpaint.

Anonymous 10/06/2022 (Thu) 23:43:52 No. 4044

>>4038 Make sure to share these and any similar on >>>/fur/

Anonymous 10/06/2022 (Thu) 23:54:12 No. 4045

>>4018 I would all 4. Has anyone tried generating a Sorayama loli yet?

Anonymous 10/07/2022 (Fri) 14:45:29 No. 4049

>>4022 What prompt did you use to get that shading? Especially on the elf.

Anonymous 10/07/2022 (Fri) 20:56:53 No. 4050

>>4043 Actually, scratch that, the waifu model 1.3 is actually great for lolis with blended art styles if you specify multiple artists with a similar style at the end of the prompt. The AI will sort of figure it out on its own and mix them.

Anonymous 10/08/2022 (Sat) 02:20:23 No. 4051

Did anybody order nightmare steampunk CLANG? >>4045 pic 5 is what the AI considers a Sorayama loli. The waifu model really loves bling and super sentai it would seem.

Anonymous 10/08/2022 (Sat) 13:57:51 No. 4052

Anonymous 10/08/2022 (Sat) 20:30:43 No. 4053

Anonymous 10/09/2022 (Sun) 02:19:51 No. 4054

This AI doesn't understand guns at all.

Anonymous 10/10/2022 (Mon) 01:08:21 No. 4056

Anonymous 10/10/2022 (Mon) 01:32:31 No. 4059

I've died and gone to discount heaven.

Anonymous 10/10/2022 (Mon) 13:40:26 No. 4060

Anonymous 10/10/2022 (Mon) 21:35:38 No. 4061

Anonymous 10/11/2022 (Tue) 00:57:43 No. 4062

>>4004 That's a man, with a loli head slapped on it.

Anonymous 10/11/2022 (Tue) 12:04:51 No. 4063

shh dont tell cuckchan about this i bet /trash/ would lose its shit if they find out about our little AI pet project despite fapping to cub themselves

Anonymous 10/11/2022 (Tue) 22:23:11 No. 4064

Making an "artist cocktail" of sorts seems to work really well with the waifu model 1.3 Basically, go to this page: https://rentry.org/artists_sd-v1-4 Grab four artists with a similar art style that you want and add them to the end of your prompt. But also use a prompt that matches the art style, like for example using a sci-fi prompt with this classic painting art style wouldn't work very well. I made these pics by adding these artist names to the prompt: Albert Lynch, Sophie Anderson, Daniel Ridgway Knight, Jean-Baptiste Monge

Anonymous 10/12/2022 (Wed) 01:13:01 No. 4065

>>4064 How much editing did you do on the hands? Or is it getting better at them?

Anonymous 10/12/2022 (Wed) 01:45:03 No. 4066

>>4065 I just use inpaint to reroll them. When the hands are in a position where not all the fingers are visible, it usually only takes a couple of rerolls to get something adequate, Sometimes you reroll, get something with the right shape but the wrong shading, and you can reroll again with a low denoising strength to just fix the shading. If all fingers are visible or, god forbid, the palm of the hand is facing the camera, its' a fucking nightmare, the AI will refuse to make sense of it, so in those cases I have to paint the hand in Krita and use inpaint to blend it into the rest of the image. In extreme cases I've copypasted a hand from a stock photo on top of the image and then used inpaint to blend it in. Sometimes 90% of the work of generating an image is getting the hands to a point where they don't draw too much attention for being horribly deformed. Oh, and hen using inpaint, always check the option that says original and the one for full resolution. I've often heard artfags complain abut hands being hard to draw, I guess they were more right than I ever expected, even an AI that can produce stunning images will absolutely shit the bed when it comes to hands. Oh, and footfags will have it even worse. The AI hates feet even more than hands.

Anonymous 10/12/2022 (Wed) 01:52:10 No. 4067

>>4065 >>4066 I forgot one tip: when using inpaint to reroll hands, rerolling the whole arm seems to help the AI understand that it's supposed to draw a hand at the end of it.

Anonymous 10/12/2022 (Wed) 13:15:08 No. 4068

Here is an artist cocktail that produces some Berserk style imagery, pair it with stuff like "haunted forest", "dungeon", "alien world", or something you would expect to see in Berserk: Greg Staples, Luis Royo, Stephen Hickman, Terese Nielsen This is in the waifu model 1.3

Anonymous 10/12/2022 (Wed) 18:19:40 No. 4069

I'm curious to see what a real artfag could do with stable diffusion, after using it for a while I can see the AI being used to remove the more labor-intensive parts of painting.

Anonymous 10/13/2022 (Thu) 07:42:26 No. 4070

Anonymous 10/13/2022 (Thu) 13:18:50 No. 4072

Has anybody tried NovelAI? I'm guessing it's probably better than running SD locally since it's a paid subscription service running on a supercompuer powered by jew gold, but I'd like some first hand reviews before spending money on it.

Anonymous 10/14/2022 (Fri) 15:00:42 No. 4095

Anonymous 10/15/2022 (Sat) 02:28:58 No. 4099

>>4072 yeah it's quite good, but you can run in locally via webui now.

Anonymous 10/15/2022 (Sat) 02:31:59 No. 4100

Anonymous 10/15/2022 (Sat) 23:12:14 No. 4103

>>4100 Nice, any time I try to generate anything with actual penetration I just get body horror.

Anonymous 10/16/2022 (Sun) 00:37:51 No. 4104

>>4103 Yeah the NAI model fixes that. You can also mix it with other models for unique results.

Anonymous 10/16/2022 (Sun) 00:59:35 No. 4105

>>4104 >>4100 >>4099 Please don't make the AI lolis cry.

Anonymous 10/16/2022 (Sun) 16:56:13 No. 4107

I've been working on my upscale-fu. Turns out you can do inpainting directly into higher resolution images, so I've been upscaling first with ESRGAN and then inpainting into the result to fix minor problems. I revisited some pics with this.

Anonymous 10/16/2022 (Sun) 21:38:51 No. 4108

The weird hands are a big turnoff. Until the NNs get better try either hiding them with prompts, photoshoping them or removing the extra fingers with https://cleanup.pictures/ Here's an example with the later tool, took me a minute, it's not perfect but beats having a nice picture ruined by extra fingers >>4061

Anonymous 10/16/2022 (Sun) 21:41:51 No. 4109

>>4108 wrong pic

Anonymous 10/17/2022 (Mon) 16:29:23 No. 4113

>>4108 When it comes to AI generated images, for now at least, you just gotta learn to live with weird hands. It's not always possible to leave the hand hidden or out of frame, sometimes you get lucky and get a decent hand that doesn't draw too much attention, but most of the time you're gonna get a horror show and it's gonna take a lot of work to make it look barely adequate.

Anonymous 10/19/2022 (Wed) 01:10:25 No. 4115

>>4099 MOAR CHUBBIES!

Anonymous 10/19/2022 (Wed) 16:57:03 No. 4117

>>4115

Anonymous 10/19/2022 (Wed) 20:04:15 No. 4118

I feel like this AI is making fun of me.

Anonymous 10/20/2022 (Thu) 11:52:17 No. 4119

Anonymous 10/20/2022 (Thu) 15:02:27 No. 4120

NovelAI definitely makes things easier but the results feel a little more samey.

Anonymous 10/21/2022 (Fri) 18:50:08 No. 4121

Anonymous 10/22/2022 (Sat) 13:38:37 No. 4122

Anonymous 10/22/2022 (Sat) 20:41:42 No. 4123

Hello? I'm here to deliver an order of deep-fried lolis. That'll be four bucks, baby!

Anonymous 10/23/2022 (Sun) 04:06:45 No. 4124

Anonymous 10/23/2022 (Sun) 08:51:31 No. 4125

>>4121 Could I kindly ask you for the prompts and settings for the girl with the blindfold?

Anonymous 10/23/2022 (Sun) 11:26:57 No. 4126

>>4125 NovelAI default settings, I don't have the exact same prompt anymore but I recreated it: >1girl, nsfw, loli, little girl, child, flat chest, skimpy outfit, arms up, bound wrists, bdsm, blindfold, ponytail, (((J. C. Leyendecker, Julie Bell)))

Anonymous 10/23/2022 (Sun) 18:42:17 No. 4127

>>4126 You are a hero. Thank you.

Anonymous 10/24/2022 (Mon) 17:27:38 No. 4129

Anonymous 10/24/2022 (Mon) 17:29:42 No. 4130

Anonymous 10/24/2022 (Mon) 17:36:10 No. 4131

I think I might be addicted to this.

Anonymous 10/24/2022 (Mon) 18:36:56 No. 4132

Go to NovelAI and use the following settings to unleash ancient and powerful lolimancy. Model: NAI Difussion Anime (Full) Prompt: 1girl, {{loli, little girl, child, flat chest, nsfw}}, dynamic pose, masterpiece portrait, {{{Adonna Khare, Ed Binkley, Jean Veber, Andreas Vesalius}}} Undesired content: blurry, depth of field, simple background, pubic hair, ugly Add or remove tags at your leisure to get specific things.

Anonymous 10/25/2022 (Tue) 00:38:57 No. 4133

>>4131 >>4130 >>4129 I really fucking love these space lolis. Holy shit, my fucking dick.

Anonymous 10/25/2022 (Tue) 20:40:02 No. 4135

Anonymous 10/25/2022 (Tue) 22:22:30 No. 4136

>>4133 I hear ya.

Anonymous 10/26/2022 (Wed) 02:21:14 No. 4137

>>4129 Things I never knew I wanted to see: loli in Boris Vallejo's art-style.

Anonymous 10/26/2022 (Wed) 14:45:28 No. 4140

>>4137 I discovered it by accident while messing with stable diffusion and it has become my favorite art style. Something about the use of colors makes the images feel more erotic.

Anonymous 10/27/2022 (Thu) 22:51:39 No. 4143

You guys are way ahead of the NAI stuff I post on /v/.

Anonymous 10/27/2022 (Thu) 22:52:20 No. 4144

Anonymous 10/28/2022 (Fri) 13:36:38 No. 4150

>>4143 Including artist names in the prompt makes a big difference, most of the artists from the link in this post >>4064 also work on NovelAI, though some of them work a little differently. Another thing to note is that NovelAI uses {curly brackets} for emphasis instead of parentheses.

Anonymous 10/28/2022 (Fri) 14:55:30 No. 4151

A schizo on /v/ posted these, but got himself deleted, so I'm reposting. Don't think he made them himself. Guy on delicious says he lazily takes all his pictures from AllTheFallen.

Anonymous 10/28/2022 (Fri) 19:41:00 No. 4152

>>4151 Text by the AI is like trying to read things in a dream.

Anonymous 10/29/2022 (Sat) 13:01:04 No. 4153

>>4144 Is there a tag for that eye shape and different types of bangs? I tend to have trouble getting specific eye shapes and bangs.

Anonymous 10/29/2022 (Sat) 16:56:10 No. 4155

>>4153 The only eye shape tag I know is sanpaku eyes. You can try including and excluding it to see what effect it might have.

Anonymous 10/31/2022 (Mon) 15:02:20 No. 4156

Happy spooksday.

Anonymous 11/01/2022 (Tue) 21:24:24 No. 4157

Today's menu: micro bikinis.

Anonymous 11/01/2022 (Tue) 21:25:04 No. 4158

Anonymous 11/03/2022 (Thu) 00:10:49 No. 4159

Can it do butthole/ass content?

Anonymous 11/03/2022 (Thu) 12:28:28 No. 4160

>>4159 I haven't been able to generate buttholes, but NovelAI has "ass" and "from behind" tags that seem to work pretty well.

Anonymous 11/03/2022 (Thu) 13:24:09 No. 4161

NovelAI is surprisingly good with tentacle porn.

Anonymous 11/04/2022 (Fri) 05:08:22 No. 4166

Anonymous 11/04/2022 (Fri) 15:19:23 No. 4169

>>4166 The tears look like cum.

flatsims 11/06/2022 (Sun) 14:11:45 No. 4171

>>4133 Did someone say space loli's? I love making them. What artist did those Boris Vallejo style? I love those too and i have a section on hd for those, would love to credit the author.....

flatsims 11/06/2022 (Sun) 14:14:43 No. 4172

flatsims 11/06/2022 (Sun) 14:15:37 No. 4173

flatsims 11/06/2022 (Sun) 14:17:04 No. 4174

>>4161 Great job, those are excellent, as are your other Boris style!

Anonymous 11/06/2022 (Sun) 17:54:20 No. 4175

>>4171 I wouldn't consider myself an artist, I'm just messing around with a fun AI.

Anonymous 11/06/2022 (Sun) 17:55:50 No. 4176

Anonymous 11/07/2022 (Mon) 02:00:25 No. 4179

Anonymous 11/07/2022 (Mon) 06:07:24 No. 4183

Anonymous 11/07/2022 (Mon) 18:13:52 No. 4184

Anonymous 11/08/2022 (Tue) 12:47:47 No. 4189

It was rather difficult to get the Ai to understand more Niche fetishes. In this case, pee, but it didn't do half bad!

flatsims 11/08/2022 (Tue) 16:08:10 No. 4190

>>4175 Well if you feel like dropping a name to associate with your art elmme know. until then you will 4 chan boris vallejo guy

Anonymous 11/09/2022 (Wed) 12:11:05 No. 4194

>>4132 In terms of stunning artwork, this produces near constant bangers, but it also loves to insert secondary faces and horror gremlins into the background.

Anonymous 11/09/2022 (Wed) 12:12:04 No. 4195

Anonymous 11/09/2022 (Wed) 12:12:43 No. 4196

Anonymous 11/10/2022 (Thu) 02:40:31 No. 4198

>>4179 >>4183 >>4184 I do not like these weird baby faces one bit. It's like the lips of a toddler and a big ugly 3DPD nose.

Anonymous 11/10/2022 (Thu) 02:45:44 No. 4199

>>4194 >but it also loves to insert secondary faces and horror gremlins into the background. It's clearly cursed, and those are hints of the demons inhabiting every image you make.

Anonymous 11/10/2022 (Thu) 02:48:49 No. 4200

>>4196 >Third image >That smug crown jewel Sugoi.

Anonymous 11/10/2022 (Thu) 11:14:53 No. 4201

>>4199 Even better, some people like the thrill of a high-risk fap.

Anonymous 11/10/2022 (Thu) 21:39:34 No. 4202

>>4150 I don't think Automatic1111 Web UI uses curly brackets even if you load the Novel AI model. I may be getting better. I don't know. Seems like a lot of time investment and RNG is involved.

Anonymous 11/10/2022 (Thu) 21:40:26 No. 4203

Anonymous 11/10/2022 (Thu) 21:41:54 No. 4204

Anonymous 11/10/2022 (Thu) 21:47:27 No. 4205

>>4169 The AI tends to do that with tears. You can try to avoid it by putting (cum on face) or (cum on body) in the negative prompt, but it might still do it. (sparkling eyes) can look pretty neat.

Anonymous 11/10/2022 (Thu) 21:58:12 No. 4206

>>4205 >You can try to avoid it by putting (cum on face) or (cum on body) in the negative prompt But where is the fun in that?

Anonymous 11/11/2022 (Fri) 00:28:22 No. 4208

Anonymous 11/11/2022 (Fri) 04:44:35 No. 4211

>>4205 Hat Kid looks like she's watching something burn.

Anonymous 11/11/2022 (Fri) 12:25:32 No. 4212

>>4190 I dunno, mongolian basket weaving boris?

Anonymous 11/11/2022 (Fri) 12:27:11 No. 4213

Anonymous 11/11/2022 (Fri) 12:28:17 No. 4214

Anonymous 11/11/2022 (Fri) 12:31:07 No. 4215

Anonymous 11/11/2022 (Fri) 17:38:06 No. 4217

Anonymous 11/12/2022 (Sat) 11:40:32 No. 4219

Anonymous 11/14/2022 (Mon) 06:25:30 No. 4220

>>4151 >>4171 >>4172 >>4173 >>4212 >>4213 >>4214 >>4215 More like these would be great. >>4172 The second one puts me in the mind of Roll. Apparently it's Roll as a French girl. In space. >>4215 That last one is absolutely adorable.

Anonymous 11/16/2022 (Wed) 16:18:14 No. 4227

Whoever teaches this fucking AI how to do hands properly should get a Nobel prize. This can get so frustrating at times.

Anonymous 11/22/2022 (Tue) 21:42:47 No. 4238

Anonymous 11/26/2022 (Sat) 01:47:54 No. 4242

>>4202 >1st pic That's a really nice foot. Those are hard to get with AI.

Anonymous 11/26/2022 (Sat) 06:54:51 No. 4243

requesting >loli jogging underwear_only

Anonymous 11/26/2022 (Sat) 10:29:42 No. 4244

Is SophiaLyn's work safe to save and post? I'm not familiar with any of this stuff and I don't know how to tell what its been trained on.

Anonymous 11/26/2022 (Sat) 12:39:10 No. 4250

Anonymous 11/26/2022 (Sat) 12:41:33 No. 4251

Anonymous 11/26/2022 (Sat) 12:44:45 No. 4252

This combination of artists on NovelAI produces some nice colorful lolis with a more comic book style: Barry Windsor Smith, Frank Frazetta, Yusuke Murata

Anonymous 11/26/2022 (Sat) 12:46:15 No. 4253

Anonymous 11/26/2022 (Sat) 13:15:52 No. 4254

@rouge87th on Twitter makes a bunch of this cute pigtailed loli with what I presume is NovelAI based on the hashtags. I'm assuming he touches her up manually to get decent hands and feet.

Anonymous 11/26/2022 (Sat) 13:18:59 No. 4255

>>4243 For some reason, the AI fucks up the hands especially hard in running poses. It would take a shitload of time to fix them, so this is the best I can do.

Anonymous 11/26/2022 (Sat) 13:21:46 No. 4256

>>4254 Yeah that's NovelAI. The cum tears in pic 2 and the foot in pic 4 gives it away.

Anonymous 11/26/2022 (Sat) 13:35:12 No. 4257

>>4256 Really wish he'd generate her having sex

Anonymous 11/26/2022 (Sat) 13:46:36 No. 4258

>>4257 Not a good idea. The AI is close to being able to generate full on porn, but all it can generate now is body horror.

Anonymous 11/26/2022 (Sat) 15:01:01 No. 4259

>>4258 With good prompts and manual editing that shouldn't be a problem. Dude's probably just autistic about only doing teases.

Anonymous 11/26/2022 (Sat) 19:47:31 No. 4260

>>4255 arigato

flatsims 11/27/2022 (Sun) 18:18:24 No. 4261

With sex scenes, you just need to make bigger batches select the best, maybe photoshop some stuff out, photoshop some better parts in from other pics, occasionally re-render, and voila. Worth time if you want it to come out right. A gift for all your great art.

Anonymous 11/28/2022 (Mon) 15:55:30 No. 4262

>>4261 Thanks. That's basically the same as doing inpainting with some photoshop tricks thrown in. I already do inpainting on some pics, the problem with sex scenes is that it takes a lot longer to get a decent base image, it takes even longer to inpaint and the results are still not on par with the solo pics and will still have significant defects.

Anonymous 11/30/2022 (Wed) 13:44:04 No. 4269

>>4261 You know, I followed your advice, it's not quite the same as inpainting after all. When you slightly change the prompt and make the AI generate a whole picture with slight differences, the result is a lot more cohesive than inpainting and produces parts of a pic with colors and style that matches the base pic, and you can pick up pieces of that variant and blend them back into the original pic. It actually makes a big difference in fixing difficult pics. Take these for example, the one on the left is the original, the one on the right I composed using your method. I don't know if I'm ready to compose whole sex scenes, but I'm definitely going to try now. With some photoshop wizardry, I might be able to pull it off.

Anonymous 11/30/2022 (Wed) 13:51:09 No. 4270

And some more stuff in the meantime.

Anonymous 12/01/2022 (Thu) 19:44:32 No. 4272

Anonymous 12/02/2022 (Fri) 13:41:28 No. 4274

Well, this takes a lot of work and still isn't perfect (the guy on the right is deformed and the guy in the bottom is a blob with a dick), but it's definitely doable with enough autism. A proper artfag would probably be able to do far better than this, I'm very limited by not being able to draw into the pic by hand, it's a shame artfags hate AI.

Anonymous 12/03/2022 (Sat) 05:40:30 No. 4276

>>4244 Literally who, and trained on what? You're not seriously coming to /loli/ to ask about 3DPD pedo style AI art are you? That's the only scenario in which I can fathom that the "training data" would somehow make the resultant pictures "unsafe to save or post".

Anonymous 12/03/2022 (Sat) 05:45:15 No. 4277

Requesting lolis wearing nothing but pasties. Star and heart shaped a best.

RogueModifier 12/03/2022 (Sat) 20:51:16 No. 4278

>>4277 Something like this?

Anonymous 12/04/2022 (Sun) 20:45:23 No. 4279

Anonymous 12/06/2022 (Tue) 03:56:00 No. 4282

>>4278 Sugoi. The future is now. Thanks anon.

Anonymous 12/08/2022 (Thu) 18:43:16 No. 4286

Anonymous 12/10/2022 (Sat) 21:57:39 No. 4291

Anonymous 12/10/2022 (Sat) 21:59:02 No. 4292

Anonymous 12/13/2022 (Tue) 03:36:51 No. 4294

>>4286 >>4291 >>4292 Doing God's work for all us plebs with toasters.

Anonymous 12/14/2022 (Wed) 15:02:05 No. 4296

>>4294 I actually started wagecucking a novelai subscription and only using my GPU for inpainting. I don't know why, but I can't get the same quality outputs from my GPU even using the leaked novelai model.

Anonymous 12/16/2022 (Fri) 01:49:26 No. 4299

>>4296 > When you realize that the leak was intentional and meant to give people a taste of the drugs they were selling.

Anonymous 12/17/2022 (Sat) 13:14:41 No. 4306

>>4299 I'm trying to get as many lolis out of it as possible. I won't be able to pay for this thing forever, the unlimited image generation is 25 burgershekels a month.

Anonymous 12/17/2022 (Sat) 13:16:04 No. 4307

Anonymous 12/18/2022 (Sun) 16:02:49 No. 4312

Anonymous 12/23/2022 (Fri) 12:55:22 No. 4333

Anonymous 12/23/2022 (Fri) 16:08:23 No. 4334

Anonymous 12/23/2022 (Fri) 17:09:52 No. 4335

Anonymous 12/23/2022 (Fri) 23:09:25 No. 4337

flatsims 12/26/2022 (Mon) 15:51:41 No. 4363

Merry Christmas Everyone!!

flatsims 12/26/2022 (Mon) 16:34:37 No. 4364

Anonymous 12/27/2022 (Tue) 04:43:42 No. 4365

Thought this board would be on fire considering how weak 4chan is on AI.

Anonymous 12/28/2022 (Wed) 04:43:22 No. 4367

>>4363 Those 3DPD faces are ugly as sin. >>4365 On fire in what way? Why would you ever go to cuckhan?

Anonymous 12/28/2022 (Wed) 20:08:47 No. 4369

>>4367 >On fire in what way? I'm assuming he means 500 pph, constant posting of AI lolis with a new thread hitting post limit every couple of hours. The problem is that, to get that kind of activity, people need to know this board exists.

Anonymous 12/29/2022 (Thu) 03:14:13 No. 4370

>>4369 >to get that kind of activity, people need to know this board exists. Even /v/ AI thread is about as slow. The whole site could be using loli, and it wouldn't make a large difference. Not enough people have enough computer, so you need a really large pool to get a tiny fraction of people to keep the thread active. Same as with regular drawfags. Love your work by the way. Will you be uploading it the newer, safer, loliproofed ourobooru?

Anonymous 12/29/2022 (Thu) 18:56:49 No. 4371

>>4370 I dunno, I don't want to get yelled at for uploading AI stuff to a booru site. But if there's a pic you like, feel free to upload it wherever and share it.

Anonymous 12/30/2022 (Fri) 02:30:50 No. 4373

>>4371 You could always tag it as AI generated. There's not much on the new ourobooru.art right now anyways. It's dedicated specifically to things made by 8anons, no matter how shitty, so as long as it's tagged as AI_generated, nobody ought to complain.

Anonymous 12/31/2022 (Sat) 02:48:53 No. 4384

>>4373 Eh, maybe I'll try uploading just a few pics later to see how it goes.

Anonymous 12/31/2022 (Sat) 03:32:36 No. 4385

>>4384 Considering how non-standard your stuff is among AI art, I expect it'll go fine.

Anonymous 12/31/2022 (Sat) 14:14:22 No. 4387

>>4385 Alright, I uploaded a few pics. If nothing bad happens after a few days I'll keep uploading the rest.

Anonymous 12/31/2022 (Sat) 14:22:34 No. 4388

>>4373 >>4387 Oh, I'm retarded, I thought you were talking about the normal ourobooru. I didn't notice ourobooru.art was a different site. I'll upload some there too.

Anonymous 12/31/2022 (Sat) 14:24:01 No. 4389

>>4388 >>4387 ~~You uploaded to the old site. Ourobooru.booru.org is at risk of take down at any moment if reported enough because it runs as part of the booru project, which bans loli/shota and other fringe content.~~ The new site is https://ourobooru.art. Make sure you use https for a secure connection or it will 404. I don't see any need to be so worried about it. Your shit is high quality. Miles above most other AI generate shit dumped to boorus en masse. However, the site has gone down for some reason.

Anonymous 12/31/2022 (Sat) 14:27:36 No. 4390

>>4388 I think it's good that's uploaded somewhere else than just here, where it might disseminate a bit more. Everything you do is top quality, but I have particular tastes and only keep my very favorites.

Anonymous 01/01/2023 (Sun) 23:55:58 No. 4401

Happy new year.

Anonymous 01/03/2023 (Tue) 20:52:41 No. 4404

>>4389 Looks like my posts were deleted from ourobooru.art, I'm guessing because there was nobody to approve them. I'll wait for the site to get more stable and have some moderators doing approvals. In the meantime, as I said, if you see a pic you like, feel free to share it wherever.

Anonymous 01/03/2023 (Tue) 23:04:07 No. 4405

>>4404 >Unapproved in three days (about 8 hours ago) Looks like that's it. I guess what's on there are only test posts by the anons fixing the site up themselves, and everyone will have to wait until Hyrdrus anon does his big dump of the old booru.

Anonymous 01/05/2023 (Thu) 17:45:37 No. 4408

Anonymous 01/06/2023 (Fri) 14:13:06 No. 4412

Anonymous 01/09/2023 (Mon) 18:12:04 No. 4421

The new ourobooru sure is taking a while.

Anonymous 01/10/2023 (Tue) 04:19:06 No. 4424

>>4421 Technical issues abounds. The software surrounding boorus is all shit and poorly documented. Many things hard coded in that shouldn't be, strange default settings like anything tagged furry being hidden by default, which must then be changed if you do something like start a furry booru. On top of that, there's concerns about server space and allowing video uploads.

Anonymous 01/10/2023 (Tue) 11:48:25 No. 4428

>>4367 They're not 3DPD looking. They're more like something out of old advertisements.

Anonymous 01/10/2023 (Tue) 12:58:01 No. 4429

>>4428 Those are 3DPD looking too. I've seen the old advertisement thread on /delicious/. It's pure boner killer.

Anonymous 01/13/2023 (Fri) 12:43:15 No. 4434

Anonymous 01/16/2023 (Mon) 13:24:48 No. 4438

I tried using the AnythingV3 model with Automatic1111 GUI but I still get lower quality outputs than what I get from novelai. Also when inpainting sometimes the AI just outputs black images, this didn't used to happen, I'm a little worried my GPU might be in trouble.

Anonymous 01/16/2023 (Mon) 16:38:52 No. 4444

>>4438 >Also when inpainting sometimes the AI just outputs black images I used to randomly get some images coming up black when I did any image generation. After some research it turned out the cause was half-precision floats. Try adding some mix of these to your command line arguments: --precision full --no-half --no-half-vae Alternatively check and see if you've switched VAE and forgot to switch back.

Anonymous 01/18/2023 (Wed) 13:00:24 No. 4448

>>4444 --no-half-vae seems to have done the trick. I didn't get any more black outputs so far. Now I'm using AnythingV3 for inpainting, it seems to be better at adding detail to eyes in particular. Still no better at fixing hands though, I hate hands. My NovelAI subscription will run out soon, I'll see what I can do to maintain the quality.

flatsims 01/18/2023 (Wed) 13:47:34 No. 4449

There's a "bad prompt v2" embedding that you put in negative prompts that is supposed to fix most hand issues. Its still the worst thing to deal with in AI. It's linked on ATF but not sure where it came from originally. Cetvai and hugging face have it possibly.

flatsims 01/18/2023 (Wed) 13:49:10 No. 4450

>>4448 Do you not run this locally on your own system? All kinds of models and adaptions should be out and freely available.

Anonymous 01/18/2023 (Wed) 15:20:55 No. 4451

I've been upscaling and retouching some old pics generated with the previous models. >>4450 I only use my GPU for inpainting because even with the latest models I can't get the same quality outputs I get from NAI. >>4449 This is the neg prompt I use: >lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name

Anonymous 01/18/2023 (Wed) 15:41:09 No. 4452

More upscales.

Anonymous 01/22/2023 (Sun) 00:05:45 No. 4456

After upgrading the Automatic1111 GUI to try some new models, I've been banging my head against the wall because of a new annoying problem with inpainting. For some reason everything I inpainted was getting slightly discolored. If anyone else is having the same problem, you have to go to the settings tab, check an option called color correction and click apply changes at the top. Though even when the colors match, I'm having trouble matching the texture of the original pic. Sometimes you get these grainy backgrounds, and when you inpaint to fix something, the AI removes the graininess from the part you inpainted, making it very obvious where the picture was altered. With the first pic here, I tried to manually add some graininess back around the hands, but it still doesn't look right. I guess I'll just try to avoid grainy backgrounds in the future.

Anonymous 01/24/2023 (Tue) 17:58:43 No. 4459

I find that adding "Bob Byerley" and "Sophie Gengembre Anderson" to the prompt generates cuter faces.

Anonymous 01/27/2023 (Fri) 12:20:13 No. 4468

I feel like I'm starting to repeat ideas. I'm willing to take some suggestions for things like hairstyles, clothes, accessories, maybe poses if they have a tag that the AI can recognize.

RogueModifier 01/27/2023 (Fri) 18:29:54 No. 4469

>>4468 >I feel like I'm starting to repeat ideas. You've done quite a lot pics, and of a wide variety, so I can understand if it seems like you're repeating stuff, if you have trouble thinking of ideas you haven't done before. >I'm willing to take some suggestions for things like hairstyles, clothes, accessories, maybe poses if they have a tag that the AI can recognize. Below is a small grab bag of ideas to try. I've tried to avoid things you've already done. Feel free to use as many or as few as you see fit. <Outfits. Try doing a girl being wrapped up in ribbons like a birthday present Use "snakeskin" as an adjective for any clothing item. (Maybe couple it with "skintight" too.) <Camera angle. Use "from above", "from below", "from side", and "from behind" to get more shots that aren't simple front on view. <Anatomy. "horns" and "bat wings" or "dragon wings" to try and get succubii or dragon-girls. "sheep ears" and "white curly hair" to get sheep-girls. "slimegirl" Statue, with the materials like gold, silver, stone and marble. <Hair. Try and get hair styles that don't have normal tag or name: Put the appropriate hair length tag in the positive prompt, and then put then put the hair styles with known tags in the negative prompt. <Accessories. Try something with blasters. I find the AI is better at doing "blaster" compared "gun" (although its still not perfect). <Specific pic ideas. A girl sitting in an imposing chair petting a cat, looking like some Bond villain. <Misc. Multiple characters. You seem to do a single character for all you pics, try putting two or three girls into some pics.

Anonymous 01/27/2023 (Fri) 18:50:03 No. 4470

>>4469 Thanks, I'll try some of those later. >Multiple characters. You seem to do a single character for all you pics, try putting two or three girls into some pics. This one is scary, it exponentially increases the chances of the AI fucking up and thus the total number of hands that I will have to fix. Fixing hands is agony.

Anonymous 01/28/2023 (Sat) 01:55:10 No. 4472

I made a quick example of how I'm making these as of now, in case anyone is interested.

RogueModifier 01/28/2023 (Sat) 05:24:00 No. 4473

>>4470 >Fixing hands is agony. Yeah, hands can be a pain to get right sometimes. My usually strategy for fixing hands is: 1. Try a simple re-roll to get better hands. 2. If that fails edit the hand to be less wrong and roll again. I find this makes it more likely to roll good hands, and you can use less de-noising strength. 3. If that fails then the hand might be too small or fine detailed. So I try and use the "Inpaint at full resolution" feature. This effectively zooms in on to just the region you're trying to inpaint (including some padding). The AI has an easier time with hands that are zoomed in or otherwise larger in the picture. >>Multiple characters. >This one is scary, it exponentially increases the chances of the AI fucking up and thus the total number of hands that I will have to fix. I do think you should brave the risks, and at least try experimenting with it for variety.

Anonymous 01/29/2023 (Sun) 04:43:55 No. 4478

>>4468 Try a girl wearing jeans or jean shorts with an open fly that just barely has her slit peeking out. Try out the "gris swimsuit". It's a meme fetish outfit like the backless sweater, so the AI should be able to identify it. Maybe some high or side angle shots where a girl with slightly loose clothes has her nipples showing without her knowing. >>4470 >Third pic There has to be some method to get the AI to show both sweat/tears/precum and cum in the same image as two clearly different fluids. These cum tears/sweat drops look really weird. >>4472 >That final inpainting step I'm surprised you got rid of her smile. I was really expecting her to be grinning with fangs at the end.

Anonymous 02/01/2023 (Wed) 07:03:09 No. 4481

>>4261 >A gift for all your great art. I looked too closely at the testicles Also what happened with her left nipple?

Anonymous 02/03/2023 (Fri) 15:25:17 No. 4486

>>4478 >I'm surprised you got rid of her smile. I was really expecting her to be grinning with fangs at the end. That was the idea at first, but it was a quick pic so I went with the first inpaint that looked cute. I usually inpaint the face, hands and ears to add detail and fix deformities. That last inpaint was mostly to add detail to the eyes and nose but it also changed her mouth and looked cute anyway so I kept it.

Anonymous 02/03/2023 (Fri) 16:43:50 No. 4487

>>4486 Good shit.

Anonymous 02/05/2023 (Sun) 23:31:29 No. 4490

This batch was a nightmare to make, you can actually see the point where I gave up.

Anonymous 02/10/2023 (Fri) 20:12:14 No. 4504

Anonymous 02/11/2023 (Sat) 23:56:01 No. 4512

flatsims 02/13/2023 (Mon) 21:35:47 No. 4521

Have some monster girls

flatsims 02/13/2023 (Mon) 21:36:49 No. 4522

a few more

flatsims 02/13/2023 (Mon) 21:38:16 No. 4523

Rarely see HC here. Have some new ones

flatsims 02/13/2023 (Mon) 21:38:54 No. 4524

Last bit

Anonymous 02/14/2023 (Tue) 18:08:22 No. 4526

>>4523 >>4524 Is this a new model? All I usually get when prompting sex scenes is body horror that would take too much effort to fix.

flatsims 02/19/2023 (Sun) 18:06:11 No. 4563

There's a TON of good new models out there for sex (and any age range). Bstaber, Bastard, and this one Magic are top notch and get good results on most things. The days of being limited are over.

flatsims 02/19/2023 (Sun) 18:11:18 No. 4564

and Controlnet is another game changer. you can even colorize doujin images pretty exactly.

Anonymous 02/20/2023 (Mon) 00:41:55 No. 4567

>>4563 >>4564 Where do you keep up to date with these models? The old lists I was using stopped updating and I hadn't noticed.

Anonymous 02/20/2023 (Mon) 23:19:37 No. 4574

I get all my stuff through ATF as they have a 300+ page thread there and plenty of model sharing. Citvai is getting pretty decent too.

Anonymous 02/24/2023 (Fri) 03:55:07 No. 4578

>>4564 That's a lot more than just colorization. Especially the first picture. Cute loli tuned into fucking el goblino niño.

Anonymous 02/27/2023 (Mon) 00:58:26 No. 4596

>>4564 I like how any written words regardless of language ends up as Violet Evergarden runes.

Anonymous 02/27/2023 (Mon) 11:37:03 No. 4600

So I've been practicing my offline generations with the newer models and tools to break free from NovelAI. I feel like my power level has increased.

Anonymous 02/27/2023 (Mon) 15:28:27 No. 4602

It's a shame that this doesn't have bones for fingers and doesn't seem to support inpainting. If it did, it would be incredibly useful for fixing hands and feet.

Anonymous 03/01/2023 (Wed) 03:40:52 No. 4616

>>4600 >>4602 Your powerlevel was already pretty fucking high, my man. How much higher could you possibly go?

Anonymous 03/01/2023 (Wed) 12:47:34 No. 4621

>>4616 Ideally until I can make sex scenes without body horror. With the openpose extension, you can get the positions of the bodies pretty well aligned, but every place where two body parts come in contact will turn into a misshapen tumor. For example, take the last pic here, you have no idea how much pain went into that one hand in the middle. After rerolling it for like an hour, I straight up grabbed a photo of a hand, pasted it over the pic, gradually deleted the old hand from underneath by painting manually over it with color from around that area, and then I had to gradually blend the photo hand in by manually balancing the colors, manually painting parts of it and then rerolling inpaints until it looked like part of the pic. And that's just for one fucking hand. You'll almost always have multiple hands doing a lot of grabbing in a sex scene, and finding a photo of a hand in the specific pose you need is not always possible.

Anonymous 03/01/2023 (Wed) 13:37:42 No. 4622

>>4578 It all depends on the preprocessor and prompt, here's my quick attempt using canny as a preprocessor. You can also preview and edit your preprocessor image. You could do things like remove text and censorship since black and white sketches are much easier to edit.

Anonymous 03/01/2023 (Wed) 17:09:54 No. 4623

The hed model for controlnet is actually great for reworking pics. These are some old pics I made with SD1.4 that I reworked with controlnet and orange mix. I hadn't posted some of these because I couldn't fix them until now.

Anonymous 03/02/2023 (Thu) 20:23:07 No. 4633

Anonymous 03/03/2023 (Fri) 00:00:49 No. 4634

>>4633 Das it mane.

Anonymous 03/03/2023 (Fri) 02:07:04 No. 4637

>>4002 It is a goddamn marvel of unrivalled proportions I pity the poor cunts without a card to run it, they need to devote their lives to getting one ASAP if they know what's good for them.

Anonymous 03/03/2023 (Fri) 02:11:52 No. 4638

>>4468 Settings and theme are important, public pool, out in the street, on a concert stage, Xmas party where the presents are dildos? All one need do is noodle on one's perversions for a bit.

Anonymous 03/03/2023 (Fri) 02:15:45 No. 4639

>>4600 So the poser thing does work offline? Excellent I'll be nabbing that pronto.

Anonymous 03/03/2023 (Fri) 13:02:42 No. 4653

Now that I have controlnet working, if anybody has a sketch or black and white drawing that they'd like to see colorized, feel free to post it here and I'll run it through my setup.

Anonymous 03/03/2023 (Fri) 21:36:41 No. 4655

>>4653

Anonymous 03/03/2023 (Fri) 23:05:25 No. 4656

(27.23 KB 1804x353 574d591b81db6b01d1032fada0d814bbdfe347bdc395e724cf5fb4e97b5ff765.png)

Cross posting this from /v/: LADS, just in case yous weren't aware SD has a temp folder where it stores a copy of every fucking image you've made (found mine last night and 12k pics sat in there) it defaults to your C:\Users\[Username]\AppData\Local\Temp see picrelated at the bottom of Saving Images/Grids in Settings to set it elsewhere and save your NvMe getting raep'd with frivolous writing.

Anonymous 03/04/2023 (Sat) 01:30:20 No. 4658

>>4655 The AI got pretty confused with those sketches. I'm guessing something with more detail to start with would work better. Or coloring the sketches first and then running them through img2img instead of controlnet. >>4656 Thanks for the heads up anon, I had a few thousand pics there.

Anonymous 03/04/2023 (Sat) 01:54:45 No. 4659

>>4658 Absolutely marvelous.

Anonymous 03/04/2023 (Sat) 02:32:44 No. 4661

>>4658 I have no idea what the fuck I'm doing.

Anonymous 03/04/2023 (Sat) 21:50:14 No. 4671

>>4574 Sorry for asking to be spoonfed but what is atf/citvai?

Anonymous 03/04/2023 (Sat) 22:10:13 No. 4672

>>4671 Allthefallen.moe. I think citvai is new ai model.

Anonymous 03/05/2023 (Sun) 12:02:18 No. 4675

Anonymous 03/05/2023 (Sun) 18:43:21 No. 4677

Any tips on how to make my generations have less weird eyes?

Anonymous 03/05/2023 (Sun) 19:14:54 No. 4678

>>4677 Depends on what you mean by weird eyes. If it's grainy eyes with low detail, that's solved by inpainting. I usually inpaint the eyes and face at 0.25 denoising strength to add detail, just remeber to select the "only masked" option (also called "inpaint at full resolution" in some versions of the UI).

Anonymous 03/05/2023 (Sun) 19:31:50 No. 4679

>>4678 Yes my eyes kind of look grainy. I'll try inpainting.

PutItInOniichan 03/05/2023 (Sun) 20:50:39 No. 4680

so ai ones allowed here or not ?

Anonymous 03/05/2023 (Sun) 20:51:29 No. 4681

>>4680 yes, see >>4000

so since it allowed i show my other work with stable i did PutItInOniichan 03/05/2023 (Sun) 20:55:09 No. 4682

here ya go and if anyone wanna help me or give me any tips here is my session 05f0a6dbddc2caab5119aa7ec10a3096cbd8f0b890ce72815410b4d80e59f0ff74

Anonymous 03/06/2023 (Mon) 14:40:09 No. 4688

Anonymous 03/07/2023 (Tue) 23:30:51 No. 4692

There's a new pose editor for automatic1111 that works better than the old openpose editor. This one is called posex, it works in full 3D, lets you save and reload poses and tells you the name of the joint you're moving. It also supports proper elliptical bones unlike the straight lines from openpose editor. Here´s the link: https://github.com/hnmr293/posex

Anonymous 03/07/2023 (Tue) 23:34:36 No. 4693

Anonymous 03/08/2023 (Wed) 00:01:28 No. 4694

>>4675 This is HEAVENLY

Anonymous 03/08/2023 (Wed) 00:06:07 No. 4695

Requesting: Loli with cow print stockings and arm gloves and is otherwise nude, lactation and/or c-cup or larger breasts would be a big plus.

Anonymous 03/08/2023 (Wed) 02:51:00 No. 4696

>>4693 Smaller. SMALLER!

Anonymous 03/08/2023 (Wed) 11:39:41 No. 4698

>>4695 This one is a lot trickier than I expected, I couldn't get a result that I was really happy with. The problem is giving the girl big breasts while still making her look like a loli. I thought I could make a pose with loli proportions and just use big breasts in the prompt, but no matter how much I shortened her torso and made her head bigger, the AI interprets that as the girl leaning forward instead of the girl being shorter. This is the best I could manage after a lot of attempts.

Anonymous 03/08/2023 (Wed) 16:13:23 No. 4699

>>4693 Is it a certain model you're using to achieve this soft/fuzzy look to your gens? Or is it a tag? I want to make this kind of look over my pictures.

Anonymous 03/08/2023 (Wed) 18:15:47 No. 4700

>>4699 It's a combination of my whole process really. The model I use right now is AbyssOrangeMix2_nsfw: https://huggingface.co/WarriorMama777/OrangeMixs But by default the outputs don't look like the ones I'm posting, that comes from various style tags that I've been experimenting with, usually a cocktail of four artist names and the AI combines their styles. This is my styles.csv right now, but I'm always experimenting with new tags: name,prompt,negative_prompt Pulp,"(Boris Vallejo, Bob Byerley, Yusuke Murata, Frank Frazetta:1.4)","" Pulp Variant,"(Boris Vallejo, Julie Bell, Robert Mcginnis, Peter Andrew Jones:1.5)","(depth_of_field, blurry, bokeh:1.4)" Faun,"(Adonna Khare, Ed Binkley, Jean Veber, Kentaro Miura:1.6), (Junji Ito:1.4), (black_and_white, greyscale:1.2)","" Fairy,"(Aaron Jasinski, Cyril Rolando, Agostino Arrivabene:1.5), (colorful:1.6)","" Ornate,"(Bob Byerley, Sophie Gengembre Anderson, Ralph Horsley, Luis Royo:1.6), (ornate:1.4)","" Anime Fuzzy,"(William Holman Hunt, Katsuhiro Otomo, John William Waterhouse:1.4),(Rumiko Takahashi:1.5), [big hair]","" Wasteland,"(Wadim Kashin, Stephan Martiniere:1.6),(intricate:1.5),(Luis Royo, Emilia Wilk:1.3)","" Fantasy,"(Gustave Doré, Gerald Brom, Gabriele Dell'Otto, Adonna Khare:1.4), (ornate, intricate:1.2)","" Fantasy Pulp,"(Christophe Vacher, Bob Eggleton, Julie Bell:1.4), (shiny skin:1.2)","(depth_of_field, blurry, bokeh:1.4)" If you copy that into your styles.csv you can use the styles dropdown in the automatic1111 UI to select them, or just copy the text into your prompt. But that will give terrible outputs without the proper negative prompts. My global negative prompt is: >"bad_prompt_version2, (worst_quality, low_quality:1.4)" With bad_prompt_version2 being a hypernetwork I got from here: https://huggingface.co/datasets/Nerfgun3/bad_prompt That should give you a good base image at 512x768, but to make it look really good I use SD upscale with the "Pulp" style from above to double the resolution and, after all that is done, I inpaint some areas at 0.25 denoising strength to add more detail, usually the eyes, face, hands and feet, but also any intricate details that look too grainy if you zoom in. So to summarize: >AbyssOrangeMix2_nsfw >style tags >bad prompt >img2img SD upscale >inpaint

Anonymous 03/08/2023 (Wed) 18:21:21 No. 4701

>>4700 >hypernetwork I meant embedding.

Anonymous 03/10/2023 (Fri) 20:17:38 No. 4711

Request: Realistic 14-year-old smiling Asian girl nude with cat ears, fire-red hair, neon (glowing) cyan-colored eyes, markings on her face and body, big boobs with blue nipples, a purple pussy, spring green public hair, and holding a sci-fi weapon. As for the setting, make it inside of a supermarket with people staring at her naked body, if possible.

Anonymous 03/10/2023 (Fri) 20:21:22 No. 4712

>>4700 Could I have the catbox of the bear girl? Her design is perfect

HildaLover 03/11/2023 (Sat) 01:17:12 No. 4714

I'm looking for a model called "Cstaber" anyone have been able to save it? I would be eternally thankful with whoever have it

Anonymous 03/11/2023 (Sat) 01:20:47 No. 4715

>>4682 sorry I'm a bit noob, where should I open that? base64? it's part of an onion link?

Anonymous 03/11/2023 (Sat) 05:08:58 No. 4718

>>4633 Any way you could do a recreation of the second picture but have them nude?

Anonymous 03/11/2023 (Sat) 14:51:14 No. 4722

>>4711 This is too weirdly specific, I couldn't get any decent output out of it. With AI you sort of have to lead it in broad directions and let it decide the specifics, that is the essence of tard wrangling. >>4712 Sorry, I clean up my base gens every couple of days so I don't have the original with all the embedded data anymore. >>4718 Color balancing is a bitch, but an attempt was made.

RogueModifier 03/11/2023 (Sat) 16:48:01 No. 4725

>>4695 I had a go at this. It took me awhile to get something I felt was oppai loli, and not a loli-faced shortstack. >>4698 >This one is a lot trickier than I expected, I couldn't get a result that I was really happy with. >The problem is giving the girl big breasts while still making her look like a loli. I thought I could make a pose with loli proportions and just use big breasts in the prompt, but no matter how much I shortened her torso and made her head bigger, the AI interprets that as the girl leaning forward instead of the girl being shorter. I find this to be a general problem with having adult themes in an image; after a certain amount it starts trying to make the character an adult too. My usual bag of tricks for fixing this is: <Positive prompt: "X-year-old girl": Specify an age typically helps, and if the results look too old, then trying lower it further. "toddler girl" can sometimes work as an alternative to age. "from above": I find this the looking down at the character point of view helps them look younger. "big eyes", "baby face": Results are a bit subjective, but they can help the character look younger. <Negative prompt: "wide hips": I find this only narrows the hips by a small amount, but it does help. For this image the method I ended up using was: First generating a loli with "medium" sized breasts, and cowprint gloves and stockings. Second using img2img at moderate strength with the loop back script and LoRAs for the lactation and larger boobs. This worked for making her big boobed, while not overly affecting her loli proportions. Unfortunately the barn background I was going for was a casualty in this step. Third there was a simple img2img pass to upscale the final pic. >This is the best I could manage after a lot of attempts. I think it came out quite lovely. >>4656 >it defaults to your C:\Users\[Username]\AppData\Local\Temp For Linux users this defaults to /tmp/

Anonymous 03/11/2023 (Sat) 18:56:44 No. 4727

>>4722 >This is too weirdly specific, I couldn't get any decent output out of it. With AI you sort of have to lead it in broad directions and let it decide the specifics, that is the essence of tard wrangling. Alright, how about this: Realistic 14-year-old Asian girl nude smiling with cat ears, red hair, and cyan eyes. Maybe that will give good output.

Anonymous 03/11/2023 (Sat) 23:07:03 No. 4728

>>4727 I added back a few elements from the old prompt, like the supermarket setting, glowing neon eyes and fire-red hair.

Anonymous 03/12/2023 (Sun) 02:00:59 No. 4729

>>4728 Looks really cool! Thanks! Would you be able to turn the nipples blue like how you removed the clothes in the first image of >>4722?

Anonymous 03/12/2023 (Sun) 02:55:16 No. 4730

Two more requests: (1.) Realistic image of a nude 16-year-old girl, a nude 11-year-old girl, and a nude 6-year-old boy on a beach. The same glowing neon cyan eyes as >>4728, but this time make the hair purple with blue streaks or alternative type of blue and purple mix (or just purple if that doesn't work). Try to make any boobs and genitals not be covered. (2.) Realistic image of a nude 45-year-old mother holding her nude 3-year-old son in a busy city location (no physical contact is fine if her holding him doesn't work). Try to make the boy's ass show. They should both have blonde hair and normal-looking green eyes.

Anonymous 03/12/2023 (Sun) 03:02:19 No. 4732

>>4730 I forgot to add that all three of them for the first request should still have cat ears. Second request should just be normal humans.

Anonymous 03/12/2023 (Sun) 03:07:26 No. 4733

>>4730 >>4731 >>4732 I forgot to add that all three from the first request should be related to the ones from the second, I'm thinking second cousins. Also they should be wearing funny hats. Oh and I almost forgot that the funny hats shouldn't cover the cat ears.

Anonymous 03/12/2023 (Sun) 03:09:29 No. 4734

>>4730 I forgot to add that their nipples should be deep sea blue and be shiny but the shiny bit should be a pinkish purple. Also there should be crab people in the background for the beach one.

Anonymous 03/12/2023 (Sun) 06:01:18 No. 4737

I should've known if I posted an addendum that people wouldn't be able to resist. Ignore >>4733 and >>4734, they're trolls.

Anonymous 03/12/2023 (Sun) 14:09:48 No. 4738

>>4729 Easy to do but this just looks wrong to me. I made another version that changes her skin color to match the blue nipples better. It's really not my thing but at least this looks less jarring. You have some weird tastes.

Anonymous 03/12/2023 (Sun) 16:50:50 No. 4739

I've been messing around with art style loras. I might take the time to learn how to make my own, they seem to give more consistent outputs than my style prompts.

Anonymous 03/12/2023 (Sun) 19:10:48 No. 4740

>>4738 I actually like the one where you changed the skin color better than the blue nipples by themselves. Thanks. >You have some weird tastes. I like a lot of variety. From the most vanilla of things to things that seem downright weird to others. But I'm not some scat or snuff freak.

Anonymous 03/12/2023 (Sun) 19:16:22 No. 4741

>>4740 But what if the girl is a masochist and then immediately revives as a ghost who then gets off on the situation?

Anonymous 03/12/2023 (Sun) 19:39:11 No. 4742

>>4741 That's a no for me. Ghost characters are fine generally speaking though.

Anonymous 03/13/2023 (Mon) 17:13:43 No. 4747

yeah we're starting to see some loli artist loras make their way to AI. Not thru citvai of course....

Anonymous 03/13/2023 (Mon) 17:14:38 No. 4748

Anonymous 03/13/2023 (Mon) 17:15:32 No. 4749

Anonymous 03/13/2023 (Mon) 17:16:32 No. 4750

Anonymous 03/13/2023 (Mon) 17:17:37 No. 4751

Anonymous 03/13/2023 (Mon) 17:18:26 No. 4752

Anonymous 03/13/2023 (Mon) 17:19:22 No. 4753

Anonymous 03/13/2023 (Mon) 20:46:56 No. 4754

>>4752 Do you have a link for the ponsuke, sabaku, tanabe, tenako, and oyari loras?

Anonymous 03/13/2023 (Mon) 23:10:26 No. 4755

>>4748 That BlueTheBone one is impressive. Sad that he was filling a much beloved niche and now people can just crank out and clean up similar art en mass. Such is the industrialization of creativity. >>4749 > feral lemma nai You sure that's an AI image? It seems too well put together. I can't spot a single irregularity. >>4752 >Ponsuke I could feel the style just from the thumbnail. Then I opened it and noticed she has two assholes. Lmao.

Anonymous 03/14/2023 (Tue) 03:04:11 No. 4758

>>4738 I wonder if you could get the nips glow like what that one furry android artist does? >>>/fur/613

HildaLover 03/15/2023 (Wed) 19:47:08 No. 4765

>>4752 Can you share the loras?

Anonymous 03/16/2023 (Thu) 01:12:27 No. 4766

>>4751 Pic 3 : the fingertips are the only thing that looks off in the whole picture. Someone, what am I missing?

Anonymous 03/16/2023 (Thu) 02:05:04 No. 4767

>>4766 >Slight irregularity in the left most eye. >Chalk board vertical border does not line up >Desk is missing a leg >Shadow of desk leg is actually another desk leg >Desk and chalkboard are very small/the girls are very large. >The floor terminates in a corner behind the blonde loli, but the wall above her is solid and flat >Blonde loli's right (PoV left) twintail has extra hair coming from the left edge of the picture, as if her left twintail is significantly longer than the right >Blonde loli is standing away from the drawer according to the floor, but her hair is hitting it as if she's right beside it >Not sure black hair loli has nipples To name a few.

Anonymous 03/17/2023 (Fri) 14:33:05 No. 4770

Anonymous 03/19/2023 (Sun) 03:59:22 No. 4774

Anonymous 03/22/2023 (Wed) 00:45:57 No. 4783

>>4774 I want to bury my face in her soft buns.

Anonymous 03/22/2023 (Wed) 13:41:16 No. 4786

MaXT 03/22/2023 (Wed) 15:23:54 No. 4788

>>4070 Could i ask what prompt and model have you used? thx

MaXT 03/22/2023 (Wed) 15:24:52 No. 4789

>>4774 What model have you used and what prompt? >>4783 >>4786

MaXT 03/22/2023 (Wed) 15:38:26 No. 4790

>>4621 My man may i ask you what model, lora, textual inversion and prompt have you used?

MaXT 03/22/2023 (Wed) 15:40:22 No. 4791

>>4600 How can you make the controlnet models look so good usually when i use it it all gets kinda strange even the hands

Anonymous 03/22/2023 (Wed) 17:06:56 No. 4792

>>4790 >>4791 It's a multi-step process. I explained it in this post >>4700 I'm using AbyssOrangeMix2_nsfw with a bunch of style tags and a negative prompt embedding to generate a base image, then I upscale it with the SD Upscale script and inpaint whatever part of the pic looks like it could use more detail. Here is a quick example using openpose. First the pose, then the base image and then the upscaled+inpainted image. (little girl, loli:1.4), (flat_chest:1.3), nsfw, (Boris Vallejo, Bob Byerley, Yusuke Murata, Frank Frazetta:1.2), topless, Negative prompt: bad_prompt_version2, (worst quality, low quality, pubic_hair:1.4), (animal_ears, pointy_ears, horns, tail:1.4) Steps: 28, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 2371987583, Size: 512x768, Model hash: 0873291ac5, ControlNet Enabled: True, ControlNet Module: none, ControlNet Model: control_openpose-fp16 [9ca67cc5], ControlNet Weight: 1, ControlNet Guidance Start: 0, ControlNet Guidance End: 1

MaXT 03/22/2023 (Wed) 17:08:42 No. 4793

>>4792 so if you need to redraw the hands what do you put on img2img

Anonymous 03/22/2023 (Wed) 17:12:20 No. 4794

>>4793 In this case the hand started out quite good, so I just inpainted it with the same prompt at 0.25 denoising strength to add a little detail. In cases where I need to fix deformed monster hands, I usually open the pic in Krita, manually paint over the deformed hand by using colors from around that area, and then inpaint the result at low denoising strengths until it's blended in, usually in the 0.25-0.4 range, keeping the same prompt as the base pic.

Anonymous 03/22/2023 (Wed) 17:28:28 No. 4795

>>4792 >>4794 Oh and here are the parameters for the SD upscale: Denoising strength: 0.25, Mask blur: 4, SD upscale overlap: 128, SD upscale upscaler: ESRGAN_4x Denoising strength and overlap are especially important for getting a good upscale.

Anonymous 03/22/2023 (Wed) 21:50:40 No. 4796

Anonymous 03/24/2023 (Fri) 11:50:23 No. 4797

Anonymous 03/24/2023 (Fri) 19:32:44 No. 4800

Anonymous 03/26/2023 (Sun) 14:14:12 No. 4808

Anonymous 03/26/2023 (Sun) 15:36:12 No. 4809

If someone had churned out so much high quality art in such a short time a few years ago, I'd question whether they were human. Truly, the future is now.

Anonymous 03/26/2023 (Sun) 16:37:50 No. 4813

>>4809 I can't wait for AI to improve in generating sex scenes and video. That will unlock infinite power.

Anonymous 03/27/2023 (Mon) 13:41:15 No. 4816

Because I'm retarded, I only now found out about a few quirks of the automatic1111 UI. Highlighting a word and pressing ctrl+up or down automatically increases and decreases emphasis, so you don't have to add all the parentheses manually. You can also make the AI ignore certain prompts during specific steps: "[prompt:number]": this ignores the prompt until the number of steps is reached. It's useful for adding details at the end of the generation when the overall image is mostly composed, like ornate details, color highlights, etc. "[prompt::number]" same as above except this ignores the prompt after the number of steps is reached. Useful for specifying broad characteristics of the image without having that prompt interfere in the later steps.

Anonymous 03/28/2023 (Tue) 02:47:04 No. 4817

>>4816 UUUUUOOOOOOOOOOOOOOOOOOOHHHHHHHHHHHHHHHHHHHH!!!!!!!!!!!!!!!! TECHNOLOGY! LONG LINES OF CODE, EROTIC!!!!

Anonymous 03/29/2023 (Wed) 14:19:32 No. 4822

Doing the base gen with cardos and upscaling with AbyssOrangeMix3 is giving me some really nice lolis.

Anonymous 03/29/2023 (Wed) 19:12:48 No. 4823

>>4822 Damn fine results, I really have to start getting into this. If I used that poser utility and then layered some word bubbles on top I could make my own doujins.

Anonymous 03/29/2023 (Wed) 19:44:07 No. 4824

>>4823 The main difficulty with doing something like doujins would be consistency. You'd need something like a specifically trained lora or else the characters would have inconsistencies in every panel, stuff like having earrings in one panel and no earrings in the next, or having elbow gloves that turn into short gloves in the next panel. Maintaining a consistent facial structure would also be a pain. That's why a proper artfag using the AI as an assistant would probably be the best for making more complex stuff like that, since he could manually draw into the gens to maintain consistency.

Anonymous 03/29/2023 (Wed) 23:54:18 No. 4825

>>4824 Couldn't this be solved by generating a single good example then using heavily weighted in-painting on that example to maintain design cohesion and using the aforementioned poser to "manipulate" the stance of the character in the output? You might wind up having to generate each character separately and then lay them out on the page through something like gimp but I think it could work. That being said, once you reach this amount of effort you have to start wondering if you would not just be better served by learning to draw (barring of course you suffer from a handicap that prevents developing the required fine motor skills such as "Stupid Fingers").

Anonymous 03/30/2023 (Thu) 10:59:16 No. 4831

>>4825 Even if you learned to draw at the master level you'd need to match good AI outputs, you'd never be able to draw fast enough to match AI speed. Generating AI images and touching them up manually will always be orders of magnitude faster than drawing the whole thing by hand, no matter how good you are at drawing.

Anonymous 03/30/2023 (Thu) 13:04:56 No. 4833

Anonymous 03/30/2023 (Thu) 23:12:44 No. 4834

Personally I'd love to see more stuff of loli Sawamura Haruka from the first Yakuza game she was in.

Anonymous 03/31/2023 (Fri) 23:15:22 No. 4838

Man, I'm hoping someone trains it to make some 2011 Wilykit

Anonymous 04/01/2023 (Sat) 06:23:57 No. 4841

Is there an AI voice model that can do loli voices? And also, can any of the leaked LLMs do loli rp or smut writing?

flatsims 04/01/2023 (Sat) 23:17:34 No. 4842

a couple new loras

Anonymous 04/03/2023 (Mon) 14:32:53 No. 4845

More lolis.

Anonymous 04/05/2023 (Wed) 14:35:04 No. 4849

Anonymous 04/09/2023 (Sun) 15:03:59 No. 4857

>>4845 >3rd pic >Master of the Killer Fist Reincarnated as a Blonde Loli I would watch that.

Anonymous 04/11/2023 (Tue) 23:53:33 No. 4901

Anonymous 04/12/2023 (Wed) 20:23:15 No. 4913

Anonymous 04/12/2023 (Wed) 20:24:20 No. 4914

Anonymous 04/12/2023 (Wed) 20:24:51 No. 4915

Anonymous 04/13/2023 (Thu) 11:40:44 No. 4917

>>4913 Love the naked overalls one. Also really like the loli warrior right after it. The way her outfit barely covers her nipples is very nice.

Anonymous 04/13/2023 (Thu) 16:12:36 No. 4918

Found an interesting model that effectively lolifies your prompt based on the strength you apply the LoRA network at. https://civitai.com/models/38551/squeezer Example images from their page. LoRA strength: 0.0, 0.5, 1.0, 2.0, 3.0

Anonymous 04/13/2023 (Thu) 16:57:03 No. 4919

>>4918 Nice. Everybody better download it before the faggots delete it from civitai.

Anonymous 04/14/2023 (Fri) 02:28:03 No. 4920

>>4915 Anon, your ability to extract what you want from the AI is art.

Anonymous 04/14/2023 (Fri) 10:48:16 No. 4923

>>4918 Loli versions of characters are 10/10. Fucking amazing. Does it only work as part of generating a new image, or can you apply it to existing images? Here's some girls I'd love to see loli versions of.

Anonymous 04/14/2023 (Fri) 12:22:51 No. 4924

>>4923 It doesn't work well enough to do that. Getting a consistent copyrighted character is not easy, you'd need a specific model or a specific lora for that character. Getting a copyrighted character with a high-detail art style is even harder.

Anonymous 04/14/2023 (Fri) 12:44:14 No. 4925

>>4924 Damn. I was hoping it had some img2img capability so it could transform rarer characters instead of just being able to build common characters from scratch.

Anonymous 04/14/2023 (Fri) 15:29:53 No. 4926

(2.30 MB 3200x740 jessica_albert-xyz_grid-0006-5.png)

>>4923 Sadly it doesn't work that well. For img2img expect it to take many liberties with your original image (unless you can make the perfect prompt for it). Additionally, if model doesn't know the character then chances are it will make the character look like someone with only vague resemblance. For example, the model knows Jessica Albert well enough and is able to do her (although with a noticeable style shift). However, it has gone and done its own thing with the background. Meanwhile Lulu isn't known to the model, and DeepBooru's reverse engineering of her appearance doesn't get close to approximating it. To do a less common character you'll need a textual inversion or LoRA for them first. <jessica albert, ?, heart, breasts, spoken question mark, nipples, jewelry, earrings, multiple boys, 1girl, bare shoulders, twintails, breasts outside, pouch, spoken heart, belt, orange hair, collarbone, blush, skirt, bag, ??, clothes pull, purple sleeves, yellow corset, red pleated skirt, standing <<lora:squeezer_:0.0> <Negative prompt: lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, blurry, sitting <Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 5, Size: 640x640, Model hash: d124fc18f0, Denoising strength: 0.6, Mask blur: 4 <transparent background, jewelry, hair ornament, necklace, 1girl, belt, breasts, full body, hair stick, lipstick, multiple belts, brown hair, bare shoulders, cleavage, solo, hair over one eye, makeup, long hair, animal on shoulder, standing, dress, holding a moogle <Lulu \(ff10\), (sharp focus:1.2), masterpiece, simple background, facing left <<lora:squeezer_:0.0> <Negative prompt: lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, pointy ears, naked, nipples, blurry, 1boy, 2girls <Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 5, Size: 512x960, Model hash: d124fc18f0, Denoising strength: 0.6, Mask blur: 4

Anonymous 04/14/2023 (Fri) 16:37:32 No. 4927

(11.53 MB 13312x868 tifa_lockheart-xyz_grid-0004-5.png)

(13.84 MB 13312x868 princess_peach-xyz_grid-0005-5.png)

>>4918 Experimented with generating a sequence, with the LoRA's weight gradually being increased. Used the popular characters Tifa Lockheart and Princess Peach. <Tifa Lockheart, (sharp focus:1.2), masterpiece, martial arts stance, red gloves, white shirt, black miniskirt, navel, from below <<lora:squeezer_:0.0> <Negative prompt: lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, pointy ears, naked, nipples, blurry <Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 5, Size: 512x768, Model hash: d124fc18f0 <Princess Peach, (sharp focus:1.2), masterpiece, wearing a pink ball gown, waving to viewer, solo, 1girl, from above, blond hair, with a castle balcony in the background, facing left <<lora:squeezer_:0.0> <Negative prompt: lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, pointy ears, naked, nipples, blurry, 1boy, 2girls <Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 5, Size: 512x768, Model hash: d124fc18f0

Anonymous 04/14/2023 (Fri) 17:45:40 No. 4928

>>4927 With ultra popular characters like Princess Peach, you don't really need the squeezer to lolify them, you can just do it with prompts.

Anonymous 04/15/2023 (Sat) 01:33:33 No. 4930

>>4923 I found specialized loras for Ryuuko, Satsuki and Jessica. I did loli Ryuuko for now. I might try the other ones tomorrow if I have some time idk.

Anonymous 04/15/2023 (Sat) 16:37:37 No. 4934

>>4930 This is amazing, though I feel that style of face doesn't suit her.

Anonymous 04/15/2023 (Sat) 16:50:06 No. 4935

>>4934 I thought so too. The cardos model makes nice proper DFC bodies but generates realistic goblina faces. Those get a little cuter and more anime after the upscale using AOM3, but I should have gone further with he anime style for Ryuuko. I'll try that later.

Anonymous 04/15/2023 (Sat) 20:53:02 No. 4936

>>4730 >>4732 So will my requests go unfulfilled? ;_;

Anonymous 04/16/2023 (Sun) 00:44:12 No. 4938

>>4930 >>4934 Loli Ryuuko version 2. I'm not entirely happy with the blurriness in the details, but for some reason the more you emphasize the character lora, the blurrier and messier the pic gets, even with inpainting, so this is the best balance I could strike. I wanted to make the face more faithful but without going completely flat 2D. This took some work so it's all I have for today. >>4936 Those requests are kinda hard and not very fun tbh.

Anonymous 04/16/2023 (Sun) 01:11:27 No. 4939

>>4938 >Those requests are kinda hard and not very fun tbh. OK, can I request (1.) a realistic image of a nude 16-year-old girl, a nude 11-year-old girl, and a nude 6-year-old boy all with with blonde hair and blue eyes, and (2.) a realistic image of a nude 45-year-old Asian woman with her 3-year-old nude son? I don't need any other specifics, they were just fun to think up. If these are still too much, can you tell me a better way to do requests?

Anonymous 04/16/2023 (Sun) 01:19:28 No. 4940

>>4939 Hair colors are not a problem, the hard part is having multiple subjects of different ages and sexes in the same pic, especially if they are interacting. It's not easy to tell the AI which subject has which characteristics. There are some extensions that ostensibly help with that like latent couple, but they are a pain in the ass to use and didn't work as well as I hoped.

Anonymous 04/16/2023 (Sun) 02:06:11 No. 4941

>>4940 OK, since one person is easier than multiple people, I'll keep it like that from now on. Can you make more like pic related? Also, more of Princess Peach, but 13 years old and naked with cyan hair?

Anonymous 04/16/2023 (Sun) 17:08:06 No. 4942

The pic in >>4941 reminded me. can you put a loli, any loli in the Gantz suit or some variant of it? I think I've seen someone else do it in AI with non-loli, so it might know the outfit. >>4938 Beautiful work. Though she's more covered up now, I like how the outfit has the suspenders she's supposed to have.

Anonymous 04/17/2023 (Mon) 00:14:35 No. 4944

Anonymous 04/17/2023 (Mon) 13:41:15 No. 4945

>>4918 Could you or someone pls pass me the file since it got deleted from CivitAI pls? >>4926 >>4927 >>4928

Anonymous 04/17/2023 (Mon) 13:45:47 No. 4946

>>4945 Yeah me too guys

Anonymous 04/17/2023 (Mon) 18:55:46 No. 4947

>>4945 >>4946 https://files.catbox.moe/s2i0yy.safetensors

Anonymous 04/18/2023 (Tue) 10:57:06 No. 4948

https://eadc6533-d76c-416a.gradio.live for a short time I will host my own GPU to render lolis for anywhom interested ;)

Anonymous 04/18/2023 (Tue) 10:59:16 No. 4949

Anonymous 04/18/2023 (Tue) 11:44:26 No. 4950

https://b9360163-a3b0-4095.gradio.live/ the program is a bit unstable and it generates a new url every time I start it. Stay under 512x512 or you'll (and I) will have a bad time

Anonymous 04/18/2023 (Tue) 15:20:10 No. 4951

>>4927 Btw what model have you used

Anonymous 04/19/2023 (Wed) 16:29:57 No. 4952

>>4951 The model used for this was AbyssOrangeMix3 ( https://civitai.com/models/9942/abyssorangemix3-aom3 ) the AOM3 version specifically (d124fc18f0).

flatsims 04/19/2023 (Wed) 23:37:13 No. 4954

Blossom

Anonymous 04/20/2023 (Thu) 15:07:55 No. 4956

Finally found some time to try the Jessica Albert and Satsuki loras. tbh the Satsuki lora is kinda bad and you never get her outfit, but she still looks cute so I'm posting it anyway. >>4941 Here you go.

Anonymous 04/20/2023 (Thu) 15:19:20 No. 4957

Does someone have a Lora or Textual Inversion for The Loud House characters or cartoons in general?

Anonymous 04/20/2023 (Thu) 18:30:44 No. 4959

>>4956 >Loli Jessica I now have one more reason to be glad for being born into this era, to witness such wonders.

Anonymous 04/20/2023 (Thu) 19:51:49 No. 4960

>>4959 Well, you're in luck, the Jessica lora is one of the better ones and works especially well when combined with the Toriyama lora.

Anonymous 04/21/2023 (Fri) 10:26:10 No. 4961

>>4960

Anonymous 04/22/2023 (Sat) 13:48:21 No. 4966

More random lolis.

Anonymous 04/22/2023 (Sat) 23:45:48 No. 4967

Anonymous 04/23/2023 (Sun) 01:55:57 No. 4968

Anonymous 04/23/2023 (Sun) 03:51:04 No. 4969

Anonymous 04/23/2023 (Sun) 04:24:49 No. 4970

Anonymous 04/28/2023 (Fri) 19:57:04 No. 4988

Do anons here upload stuff to boorus? I have a question. Can you upload a bunch of pics at the same time and just tag them as AI or do you have to add all the relevant tags one by one to each pic? Because I noticed the booru sites are pretty demanding when it comes to tagging, I have to think of like 20 tags for each image and that would take a lot of time and effort to upload.

Anonymous 04/28/2023 (Fri) 19:58:04 No. 4989

Anonymous 04/29/2023 (Sat) 12:31:52 No. 4990

>>4988 Go ask here. >>>/v/724311 They might be ready for regular uploads now.

Anonymous 04/29/2023 (Sat) 12:38:04 No. 4991

>>4989 >Gantz loli Very nice. Sad the outfit wouldn't remain a little more consistent though. May one day I'll commission an edit to fix that.

Anonymous 04/29/2023 (Sat) 17:02:37 No. 4992

>>4144 I'm a bit late to this board but I made an Ashley lora a while ago among a bunch of other loras people here might be interested in https://mega.nz/folder/2FlygZLC#ZsBLBv4Py3zLWHOejvF2EA

Anonymous 04/30/2023 (Sun) 00:39:58 No. 4993

>>4992 neat

Anonymous 04/30/2023 (Sun) 00:45:36 No. 4994

Anonymous 04/30/2023 (Sun) 01:28:11 No. 4995

Anonymous 05/02/2023 (Tue) 15:35:41 No. 4998

>>4992 Thats neat what model and prompt have you used? maybe also loras? if you used anything else like controlnet or something lemme know thx cause i never get the same results as other people even if i copy their exact settings

Anonymous 05/02/2023 (Tue) 17:07:46 No. 4999

>>4998 Here's a few I already had uploaded to catbox, just save them and check the png info in webui. https://files.catbox.moe/tmu8lx.png https://files.catbox.moe/x7fajq.png https://files.catbox.moe/dayjhf.png https://files.catbox.moe/n2ry75.png https://files.catbox.moe/f93kqq.png Generally for models I use Based64 but AOM2/AOM3 or any of its mixes will get you most of the way there. I almost always use hires fix at 2x with 4x-Ultrasharp and denoise ranging around 0.4-0.5. Thankfully it looks like I don't have to remind anyone here to download a vae but I just use ema-560000. Most of the loras I use are mine and have uploaded in the original link. Aside from that my starting prompts are pretty simple, for example: [code](best quality:1.2), nsfw, solo, nude, 1girl, loli, cowboy shot, from below,[/code] etc. [code]Neg: (worst quality:1.2)[/code] Unless I'm using a character lora in which case I usually use a BREAK and put character details on their own line.

flatsims 05/02/2023 (Tue) 18:55:00 No. 5000

A little BanG Dream! Girls Band Party!

flatsims 05/02/2023 (Tue) 18:57:21 No. 5001

And some PuypoPuyo

flatsims 05/02/2023 (Tue) 18:59:35 No. 5002

And some general guitar girls

flatsims 05/02/2023 (Tue) 19:01:21 No. 5003

Anonymous 05/05/2023 (Fri) 16:56:01 No. 5006

>>4999 So first you create the image and then denoise it? Also is vae really that important? And that [/code] thing what is it?

Anonymous 05/05/2023 (Fri) 16:56:52 No. 5007

Also you get the feet and hands so perfect only by generating the image normally? >>4999

Anonymous 05/05/2023 (Fri) 17:00:29 No. 5008

I cant find the based64 model btw

Anonymous 05/07/2023 (Sun) 12:18:47 No. 5010

>>5008 First result on DDG. https://huggingface.co/AnonymousM/Based-mixes/blob/main/Based64mix-V3.safetensors You are bad at searching, anon.

Anonymous 05/08/2023 (Mon) 16:43:36 No. 5011

>>5010 Yeah I m kinda new to those things :P

lili 05/09/2023 (Tue) 19:05:21 No. 5012

I love you dad

meta 05/11/2023 (Thu) 13:31:51 No. 5013

Reminder there's an loli-friendly AI community on session currently at 1790 users. Rules: Posting or asking for real stuff will get you banned immediately. Copy this into the "join community" section in the app. http://46.101.253.18/aiunlimited?public_key=fc309f9c04c4204cb1a62ae75b1dd9e99dd88cc3777097fb95a614952f598f0b pastebin: bericbone.com

Anonymous 05/18/2023 (Thu) 16:06:42 No. 5019

Mixing a bunch of different style Loras at low strength gives some pretty nice results.

Anonymous 05/18/2023 (Thu) 16:07:40 No. 5020

Anonymous 05/18/2023 (Thu) 16:08:32 No. 5021

Anonymous 05/19/2023 (Fri) 03:15:00 No. 5023

Request: Dahyun from Twice (K-pop group).

Anonymous 05/19/2023 (Fri) 05:52:19 No. 5024

>>5020 They say that photography killed art (and I agree that it did), but now art has died again as only having a purpose to create new styles for the machines to copy.

Anonymous 05/19/2023 (Fri) 10:38:40 No. 5027

>>5019 >>5020 >>5021 It's been awhile. Godly work as always.

Anonymous 05/20/2023 (Sat) 01:54:13 No. 5028

>>5027 Thanks, anon. I've just been addicted to vidya for a couple of weeks. Since ourobooru is taking so long I'm thinking I might upload some stuff to atf booru since it looks pretty AI friendly. Tagging every pic with 15+ tags each seems like a lot of effort though. >>5024 I don't think it will be that dramatic, once artists stop sperging out and start using AI it'll just make the process easier and more fun. With inpainting you can really have full control over the pic and you don't depend on prompts so much. I think artists could easy draw a sketch by hand, do some basic coloring, and then use inpainting to add detail, blend in new elements, etc. The AI is incredibly useful for just blending stuff into a pic, you can gradually compose anything you want like that if you just put enough autism into it. In contrast, I'm just messing around with art styles, starting from prompts and then upscaling and tweaking the result, it's nowhere near the best use of AI.

Anonymous 05/20/2023 (Sat) 12:50:54 No. 5029

>>5028 >Tagging every pic with 15+ tags each seems like a lot of effort though. That's too much for most of these solo pics of generic characters in my opinion. Off the top of my head I can think of about nine they likely have that I'm sure you could apply to most of your images. >loli >1female >solo >explicit >nude >hair length >hair style >hair color >eyes color Maybe they also have tags for poses?

Anonymous 05/20/2023 (Sat) 22:22:16 No. 5030

Anonymous 05/23/2023 (Tue) 07:03:52 No. 5036

>>4658 how are they so consistent to the original? Mine sometimes don't even have the same pose

BERBOS 05/24/2023 (Wed) 16:52:37 No. 5039

llolli

Anonymous 05/25/2023 (Thu) 04:25:26 No. 5041

Reminder that there's an AI board on wikieat.club if anyone wants to post there. We could really use the traffic!

Anonymous 05/25/2023 (Thu) 13:58:19 No. 5042

>>5030 Can you maybe suggest a website or a video that you used to learn to be so precise with Stable diffusion or to show the same process you use?

Anonymous 05/25/2023 (Thu) 19:38:39 No. 5043

>>5036 I made those with controlnet canny. Canny should follow the base sketch pretty closely, it's the same thing I used for this >>4622

Anonymous 05/25/2023 (Thu) 19:57:50 No. 5044

>>5041 The website says making updates and does not load for me

Anonymous 05/26/2023 (Fri) 12:16:46 No. 5045

>>5043 >Those sagging pants

Loliver 05/28/2023 (Sun) 16:34:27 No. 5048

Those are my first works so I hope you like them. Also pls be honest with the reviews so next time I can improve them. I'll use more posts since it wont let me post all at once.

Loliver 05/28/2023 (Sun) 16:36:33 No. 5049

Loliver 05/28/2023 (Sun) 16:37:02 No. 5050

Loliver 05/28/2023 (Sun) 16:38:04 No. 5051

Anonymous 05/29/2023 (Mon) 11:48:28 No. 5052

>>5048 Pretty solid start. If you want to improve, you should read what the high quality loli AInon posted in this thread so you can work on inpainting out some of the rough bits. Also check out the AI art thread on /v/.

Loliver 05/29/2023 (Mon) 13:33:06 No. 5053

Thx. Btw who is this AInon

Anonymous 05/30/2023 (Tue) 11:19:54 No. 5055

>>5053 Random name I made up for the guy making 95% of the top quality pics in this thread, though the name could be used for any anon using generative AI.

Anonymous 05/31/2023 (Wed) 15:03:26 No. 5058

>>5048 Neat. I'd say use inpainting to fix hands and feet and don't be afraid to manually paint into your pics because you can use inpainting to blend things in. You can inpaint directly into high resolution images if you mark the "only masked" option in the Automatic1111 UI and, at low denoising strength (like 0.25) it's very useful to add detail to important parts of the pic. For example, one simple thing you can try is to inpaint the eyes at 0.25 denoising strength to smoothen things out and make them look better. Then from that you can move on to fixing hands, feet and other details that the AI is prone to messing up. Anyway, I've been posting less because I got addicted to eraTohoK after having ignored it for the longest time. Now the good news is that I used my autism to make a bunch of portraits for custom characters, they're in this zip file: https://files.catbox.moe/ypiuu9.zip If you have those slots occupied just rename the pics. The last three portraits are male in case you want to play a conqueror to give the girls the D.

Anonymous 06/01/2023 (Thu) 18:58:52 No. 5062

This detail tweaker LoRA is really useful to make more intricate pics: https://civitai.com/models/58390/detail-tweaker-lora-lora?modelVersionId=62833

Anonymous 06/02/2023 (Fri) 10:53:28 No. 5069

>That loli Mirko Fucking 11/10, pure solid gold.

Anonymous 06/06/2023 (Tue) 15:21:19 No. 5072

>>5069 Thanks anon, that was a pretty tricky one to make because the AI got a lot of details wrong about her design that I had to fix manually. But I managed to fix most of them. Anyway, if you've noticed, my pics are looking less grainy in the last couple of posts. Turns out that grainyness my pics had is something called adversarial noise. I have no idea why I get so much noise in my gens when other anons posted images with perfectly smooth colors, but the grainy effect has been bothering me for some time. Luckily I found this extension to smoothen out my pics: https://github.com/lllyasviel/AdverseCleaner Now, the filter that applies is not perfect. It makes some areas look smudged and removes some detail from parts of the pic you'd expect to have more texture, like grass and hair, so what I've been doing is paste the cleaned up image in a new layer on top of the original and then use a low opacity soft eraser to erase the cleaned top layer from areas that could use more texture. I'm decently happy with the results, but between that and the color balancing shenanigans I've been doing, my pics are taking longer to finish these days. I wonder if I'm getting too autistic about adjustments like that.

Anonymous 06/07/2023 (Wed) 11:56:24 No. 5084

>>5072 No such thing as too much autism in the perfection of art.

Anonymous 06/08/2023 (Thu) 11:49:17 No. 5086

Can you make a loli Josuke?

Anonymous 06/09/2023 (Fri) 19:30:19 No. 5090

>>5084 Autism it is then. >>5086 Sure, the character LoRA for Josuke gives some pretty good results, you just have to emphasize the pompadour a lot or the AI will give loli Josuke long hair.

Anonymous 06/10/2023 (Sat) 13:29:37 No. 5092

>>5090 Fucking impressive.

Anonymous 06/12/2023 (Mon) 13:44:28 No. 5096

In case anyone is curious about how much of a difference upscaling + inpainting makes, here are some comparison pics between the unedited low resolution images straight out of a prompt and the finished ones. Also one new pic.

Anonymous 06/13/2023 (Tue) 10:37:04 No. 5100

>>5096 Gives a greater appreciation of how much work you put into these.

Small Slut Ammy 06/15/2023 (Thu) 19:14:14 No. 5111

Tiny girl with huge ass full nude and wet

tiny lady Waos 06/17/2023 (Sat) 06:51:10 No. 5113

twin type hair showing pussy

Anonymous 06/23/2023 (Fri) 19:57:18 No. 5118

>>5100 Thanks anon. I'm spending increasingly more time on each image as I learn more tricks and get more attentive about the details. The downside is that my posting has slowed down.

Anonymous 06/25/2023 (Sun) 05:52:54 No. 5120

>>5118 >pic 5 >haughty/snobby loli My pee pee is too hard.

Loliver 07/02/2023 (Sun) 11:14:25 No. 5140

I'd like to share this Lora of JCM2 style since it got taken down on civitai and it is now pretty rare to find it so I hope you like it. https://mega.nz/file/1fFWWZhI#293kBb-R7worKukdknpaZPoeY3N0pGbY0HyJi8IWfaY

Anonymous 07/02/2023 (Sun) 13:57:29 No. 5141

>>5140 Cool. Though I advise you that namefagging is frowned upon unless you're a drawfag, devfag, or some other kind of creator that would be recognizable anyways through your works.

Anonymous 07/05/2023 (Wed) 19:32:14 No. 5148

Using these settings in Adverse Cleaner makes the pic look less smudged after cleanup, requiring much less work with the soft eraser afterward. Bilateral Filter Diameter: 1 SigmaColor: 4 SigmaSpace: 4 Guided Filter Radius: 1 Accuracy: 8 That still cleans up grainy pics quite nicely, but also removes less of the detail that is actually part of the art style.

Anonymous 07/06/2023 (Thu) 12:13:28 No. 5149

>>5148 For the first pic, you may want to raise her right arm (viewer's left), as it seems like at that angle and lowness, that her upper arm is too long. Nice Pan.

Anonymous 07/06/2023 (Thu) 13:16:15 No. 5151

>>5149 Goddammit how did I miss that? Thanks for the heads up anon. This is a good opportunity for a little tip about manually painting into your pics: if it looks good when you squint, then the AI can blend it in perfectly at 0.25 denoising strength. Speaking of fixing stuff, this second pic is driving me crazy trying to figure out how to fix it. It looks too good to ditch, but that right foot is not actually resting against the wall... dammit.

Anonymous 07/06/2023 (Thu) 21:06:27 No. 5153

The easiest solution I could see for that right foot was to just put a box behind her, and it's a good example of that squinting rule I mentioned, It took a couple of tries but look at how perfectly the AI blended my shitty fridge tier box into the rest of the pic, you just need to paint something that's barely enough for the AI to understand what you're going for and it will quickly get it right.

Anonymous 07/07/2023 (Fri) 12:21:39 No. 5155

>>5151 Nice fix. Fukken saved. >>5153 That punk loli is pure fucking gold. Especially the wrap top with her nipple just barely showing. Whether you believe she's oblivious, or pretending to be oblivious to it, it's fucking hot.

Anonymous 07/14/2023 (Fri) 14:34:32 No. 5162

Some experiments mixing cardos and based64. >>5155 >That punk loli is pure fucking gold. Especially the wrap top with her nipple just barely showing. Whether you believe she's oblivious, or pretending to be oblivious to it, it's fucking hot. Thanks anon, that's exactly why I didn't want to abandon the pic because of the foot.

lolis IA kh4k4r0t0 lolis 07/14/2023 (Fri) 17:42:16 No. 5164

for ATFbooru

Ai gen loliss Kh4k4r0t0 lolis 07/14/2023 (Fri) 17:57:40 No. 5165

To post on ATFbooru tagged kh4k4r0t0

More lolis Kh4k4r0t0 lolis 07/14/2023 (Fri) 18:01:32 No. 5166

Kh4k4r0t0

Anonymous 07/15/2023 (Sat) 04:17:28 No. 5167

>>5165 There's already an active thread for this. Check the catalog before making threads.

Anonymous 07/15/2023 (Sat) 04:26:57 No. 5168

>>5159 This made me vomit. >>5162 >Thanks anon, that's exactly why I didn't want to abandon the pic because of the foot. I've seen a lot of AI pics where if nipples or genitals are just half covered, the AI fails to understand the positioning and shape of the pink bits in question. I take it good partial exposure is hard to get right, just like fingers and the genitals themselves? >Some experiments mixing cardos and based64. I feel an increase in realism and decrease in stylization compared to your recent pictures. I prefer stylization myself.

Anonymous 07/15/2023 (Sat) 18:44:15 No. 5169

>>5159 I've really gotten into newly-blossoming pubic hair. >>5162 I love the more realistic stuff. I hate stylized crap. Calartism is a disease that needs to be fought wherever possible.

Anonymous 07/15/2023 (Sat) 19:08:24 No. 5170

Anonymous 07/16/2023 (Sun) 02:19:28 No. 5171

Anonymous 07/16/2023 (Sun) 06:32:23 No. 5173

Anonymous 07/16/2023 (Sun) 16:29:32 No. 5174

>>5169 >I hate stylized crap. >>5168 >I prefer stylization myself. Ah, the duality of man. >I take it good partial exposure is hard to get right, just like fingers and the genitals themselves? It's pretty annoying yes, it usually requires a good amount of manual painting and tard wrangling to prevent the AI from deleting or misinterpreting the partially exposed parts.

Anonymous 07/18/2023 (Tue) 01:12:31 No. 5175

Anonymous 07/18/2023 (Tue) 03:57:37 No. 5176

Anonymous 07/21/2023 (Fri) 23:44:47 No. 5206

Anonymous 07/22/2023 (Sat) 13:44:05 No. 5207

>>5206 Pure gold as always. Shame this site is dead. Do you upload anywhere else? Maybe you should go to one of those loli friendly social medias and dump your work and/or shill this place. I think they were called Pawoo and Baraag. One of them is a Mastodon instance but I forget which.

Anonymous 07/23/2023 (Sun) 04:12:53 No. 5218

>>5207 Everywhere is dead because you cannot post anything even slightly controversial anymore.

Anonymous 07/24/2023 (Mon) 14:30:23 No. 5226

>>5207 Thanks anon. The idea was to upload to ourobooru but the site is taking forever and my posts got deleted the last time so I'm not 100% sure AI stuff is welcome. I made an account on atfbooru some time ago and I was going to start uploading there since I see a lot of AI loli stuff there and it doesn't get deleted, but I've just been lazy because of all the tagging I need to do. I finally got off my ass and uploaded some pics to atfbooru, I'll try to post my stuff there too from now on, so you should find them here tagged with "user:uohtism".

flatsims 07/24/2023 (Mon) 15:56:22 No. 5234

>>5218 The reason why 8chan suffers as a site is the same reason most chans suffer. People hide behind anonymity and act like assholes in most cases. Thje concept of don't shit where you eat is lost on a fair amount of the people here. Other places don't have this problem....

flatsims 07/24/2023 (Mon) 15:57:58 No. 5235

A little Re:Zero

flatsims 07/24/2023 (Mon) 15:58:33 No. 5236

flatsims 07/24/2023 (Mon) 15:59:09 No. 5237

Anonymous 07/24/2023 (Mon) 20:48:11 No. 5240

>>5218 Controversial? Honestly I can't think of anything more universally loved than naked girls

Anonymous 07/27/2023 (Thu) 02:14:09 No. 5285

Anonymous 07/28/2023 (Fri) 17:36:03 No. 5307

Anonymous 07/29/2023 (Sat) 13:07:32 No. 5311

>>5307 >Multi-character works Challenging yourself, I see. How long did it take to get the right (viewer's left) hand correct on that second pic?

Anonymous 07/29/2023 (Sat) 15:46:18 No. 5312

>>5311 Not as long as you'd expect. The base hand wasn't monstrously deformed and the upscaling already got 3 out of five fingers right. So I only had to manually fix two fingers.

Anonymous 08/01/2023 (Tue) 13:47:58 No. 5325

And today we have some experiments that nearly fried my GPU and reminded me of the need to upgrade. When upscaling to very large resolutions, the AI has to do it in small chunks of 768x768 or, in my case, the GPU will run out of memory. Each chunk that the AI has to enhance will be lacking the context of the whole picture, so it will try to add stuff from the prompt that should not be in that chunk. To avoid this, I've found that the DPM++ 2M SDE Karras sampler respects the base image a lot more when upscaling.

Anonymous 08/02/2023 (Wed) 10:31:55 No. 5338

>>5325 Fantastic work as always. In the second pic, is that meant to be worn that way, or is it a nip slip? In the third pic, the details on the hand might be too realistic. There's a reason even the more highly detailed anime style artists practically never draw detail on the knees, elbows, and knuckles of girls. I can say the same about the wrist of the last girl visible in the bottom left corner. Realistic wrinkling in these areas just isn't very attractive to most people who like the smooth clear skin of cartoon women.

Anonymous 08/03/2023 (Thu) 21:09:23 No. 5360

Anonymous 08/04/2023 (Fri) 01:15:56 No. 5361

Anonymous 08/04/2023 (Fri) 22:23:01 No. 5372

>>5207 >>5218 I've tried advertising this before, but wikieat.club allows loli AI nudity.

Teleo 08/06/2023 (Sun) 06:17:04 No. 5385

I just got Stable Diffusion on my machine and spent all day generating images. These are probably my top five; I mostly tried to emulate Uohtism's setup that he talked about. Any critiques? So far, I don't understand inpainting; it seems whatever the software newly generates never matches the underlying image. My favorite images from this thread is the second image from >>4335 and the second image from >>4833 Wow, goddamn gorgeous. What were the prompts, setups, and processes for those? o=

Anonymous 08/06/2023 (Sun) 14:58:30 No. 5387

Anonymous 08/06/2023 (Sun) 15:11:26 No. 5388

Anonymous 08/06/2023 (Sun) 17:02:06 No. 5389

>>5338 Thanks anon. >In the second pic, is that meant to be worn that way, or is it a nip slip? I imagine it's either a slutty tube-like leotard or it's some kind of curtain that she's pulling down with her legs to reveal her nipples. Either way, that expression should tell you she's doing it on purpose. > In the third pic, the details on the hand might be too realistic. There's a reason even the more highly detailed anime style artists practically never draw detail on the knees, elbows, and knuckles of girls. I can say the same about the wrist of the last girl visible in the bottom left corner. Realistic wrinkling in these areas just isn't very attractive to most people who like the smooth clear skin of cartoon women. I can sort of see it with the wrist. After spending so much time fixing hands, removing detail from a hand goes against my instinct, but I'll keep that in mind. >>5385 I'm happy to see more people getting into it, those are some really nice pics. >My favorite images from this thread is the second image from >>4335 and the second image from >>4833 Wow, goddamn gorgeous. What were the prompts, setups, and processes for those? o= Thanks anon. The first pic is pretty old, I generated that one with NovelAI and only inpainted it with my local SD. I don't have the prompt anymore but I'm pretty sure it had heavy emphasis on Boris Vallejo and Julie Bell. The second pic is newer but I also didn't save the prompt. In that case I'm pretty sure I used the cardos animated model and the "painterly" style preset (posted below). > Any critiques? So far, I don't understand inpainting; it seems whatever the software newly generates never matches the underlying image. The obvious one is that the pics aren't inpainted, so they tend to lack detail in the eyes. The basic thing with inpainting is to check the "only masked" option, use the same prompt as the base image (you can alter the prompt slightly to get specific things in the area you're inpainting), and generally use low denoising strength in the 0.25-0.45 range. Another important trick I've learned with inpainting is something I like to call "context points" which I try to explain in the last pic here. Basically, when inpainting only the masked area, the AI will grab a piece of your pic, upscale it, generate it again and then downscale it back into the pic. The problem is that the AI will only "see" the little square that it's inpainting and it may lack the context it needs to understand what the pic is supposed to be, so what I like to do is add small dots away from the main masked area to increase the size of the area that the AI can "see" during inpainting. A typical example of this is inpainting a hand in a weird position; by just masking the hand itself, they AI won't be able to tell that it's supposed to be a hand, so you can add a small dot on the elbow and now the AI will expand the inpainting canvas so it can see the whole forearm and it will be much more likely to understand that it's supposed to paint a hand. Anyway, since you're probably using my old styles.csv that I posted a while ago, here's my updated file that I've been using lately with more tweaks and experiments. name,prompt,negative_prompt Pulp,"(Boris Vallejo, Bob Byerley, Yusuke Murata, Frank Frazetta:1.3)", Pulp Variant,"(Boris Vallejo, Julie Bell, Robert Mcginnis, Peter Andrew Jones:1.3)","(depth_of_field, blurry, bokeh:1.2)" Space Pulp,"(Terese Nielsen, Ken Kelly, Jim Warren, Mark Brooks:1.3)", Sci-Fi Landscape,"(Bob Eggleton,Paul Lehr, Barclay Shaw, Peter Elson:1.4)", Inky,"(Philippe Caza, Adam Hughes, Alice Pasquini, Greg Staples:1.3)", Cute Grill,"(Bob Byerley:1.4), (Albert Lynch:1.4), (Sophie Gengembre Anderson:1.3), (Ralph Horsley:1.3)", Painterly,"(Alexandre Cabanel, Jean-Baptiste Monge, Albert Lynch:1.3)", Faun,"(Adonna Khare, Ed Binkley, Jean Veber, Kentaro Miura:1.4), (Junji Ito:1.2), (black_and_white, greyscale:1.2)", Fairy,"(Aaron Jasinski, Cyril Rolando, Agostino Arrivabene:1.3), (colorful:1.4)", Ornate,"(Bob Byerley, Sophie Gengembre Anderson, Aaron Horkey, Luis Royo:1.4)", Wispy,"(Tomasz Alen Kopera, Agostino Arrivabene, Neil Blevins:1.3), (Luis Royo:1.4)", Fairy Mix,"(Cyril Rolando, Agostino Arrivabene:1.3), (Luis Royo, Bob Byerley:1.4)", Anime Fuzzy,"(William Holman Hunt, Katsuhiro Otomo, John William Waterhouse:1.2),(Rumiko Takahashi:1.3), (big_hair)", Sci-Fi,"(Wadim Kashin, Stephan Martiniere:1.4),(intricate:1.2),(Luis Royo, Emilia Wilk:1.2)", MAF,"(Greg Staples, Luis Royo, Stephen Hickman, Terese Nielsen, Kentaro Miura:1.3)", Gachashit,"(Peter Mohrbacher, Raymond Swanland, Tommy Arnold, Aleksi Briclot:1.4)", High Detail,<lora:add_detail:1>, Hyper Detail,<lora:add_detail:2>, Low Detail,<lora:add_detail:-1>, Furry Begone,,"(animal_ears, pointy_ears, horns, tail:1.4)" On that note, there is one more tip of the A1111 UI: you can edit your ui-config.json file to include things like your preferred default pic size, base prompts, and all sorts of options that will be preloaded when you start the UI. I've found that this saves a lot of time. For example I have default base prompts and negative prompts that I use and inpainting options checked by default: "txt2img/Prompt/value": "(little girl|loli:1.4), (flat_chest:1.3), nsfw", "txt2img/Negative prompt/value": "bad_prompt_version2, (worst quality, low quality, pubic hair:1.4), <lora:easynegative:1>", "txt2img/Sampling method/value": "DPM++ 2M Karras", "txt2img/Sampling Steps/value": 28,

[Expand Post]

"txt2img/CFG Scale/value": 11.0, "txt2img/Width/value": 512, "txt2img/Height/value": 768, "img2img/Masking mode/value": "Inpaint masked", "img2img/Sampling Steps/value": 28, "img2img/CFG Scale/value": 6.0, "img2img/Denoising strength/value": 0.25, "customscript/sd_upscale.py/img2img/Tile overlap/value": 128, "customscript/sd_upscale.py/img2img/Scale Factor/value": 2.0,

Anonymous 08/06/2023 (Sun) 17:13:14 No. 5390

>>5389 Oh, I forgot the masked option for inpainting: "img2img/Inpaint area/value": "Only masked",

Teleo 08/07/2023 (Mon) 05:27:47 No. 5395

>>5389 Thanks for the advice, Anon. I put your inpainting tips into practice. The two images I attached show my best efforts. When I say I tried to emulate your setup, what I meant was I copy and pasted your prompts and negative prompts, and punched in the settings you mentioned in your post >>4792 for the most part. I didn't know anything about styles.csv. How does styles.csv work? And thanks for sharing your ui-config.json, it really made things easier for me. Interestingly, in my ui-config file, "img2img/Masking Mode/" is just "img2img/Mask Mode/". Slight SD update, maybe? Another question; your height and width values are 768 and 512 in your ui-config, but you commonly post images in this thread that are over 1,000 pixels both ways and high-quality. Are you using a certain upscaler after generating your images?

Anonymous 08/07/2023 (Mon) 10:48:35 No. 5397

>>5395 Amazing pics, you got it. >Another question; your height and width values are 768 and 512 in your ui-config, but you commonly post images in this thread that are over 1,000 pixels both ways and high-quality. Are you using a certain upscaler after generating your images? Yeah, I thought you had figured that out because of the 1024x1024 pics you posted, but apparently you have a much better GPU than mine since I can only go up to 768x768 without crashing half the time. Anyway, in the img2img tab, there is an option at the bottom for user scripts, select SD upscale, set the denoising strength to something low like 0.25 and the size to whatever you want your chunks to be, since your GPU can handle 1024x1024 I'd use that for the chunk size. Select an upscaler from the options below (ESRGAN 4x works well in most situations, even if you're only upscaling to 2x) then hit generate and the AI will divide the pic into chunks, upscale them and blend them into each other. You might have to do a little inpainting directly onto the upscaled image to blend everything in perfectly. Also, it seems the AI likes to divide the sampling steps between chunks, so I usually increase the sampling steps to something like 70 while upscaling. As for the styles.csv, that's a file in the root of the A1111 folder, everything you have in that file will be loaded into the styles dropdown list at the top right. They are basically parts of prompts that will be appended at the end of your main prompt during generation, so it's perfect for reusable style prompts, loras, etc.

Anonymous 08/07/2023 (Mon) 11:25:47 No. 5399

>>5389 >that expression should tell you she's doing it on purpose. That slight smile? She's obviously wearing a lewd outfit regardless of whether a nip slip is involved, so that expression could work either way. It's hot when girls are trying to tease you by dressing or behaving lewdly without actually trying to show any naughty bits, but unintentionally show them anyways. Spoilers for non-lolis.

Anonymous 08/07/2023 (Mon) 11:35:21 No. 5400

>>5399 >2 different versions of loli Bulma Is one from an OVA or something?

Teleo 08/08/2023 (Tue) 17:46:40 No. 5432

>>5397 Thanks for all the help, Anon! I find that upscaling the entire image smooths over imperfections just as well as spot-inpainting. Although it still screws up sometimes, and inpainting or upscaling has the tendency to make notable changes to faces, which is a downside if one liked a previous face more than a newer face, but those are just nitpicks. Attached is an original txt2img, then its spot-inpainted img2img that I previously posted, and then an upscale of the original txt2img. I like the upscale the most overall, but I think I prefer the face from the spot-inpainted one. I also upscaled the other pic that I posted with the two girls, but one of the hands got messed up really badly. I got some other questions. What's the hed model for ControlNet? What's Based64Mix? What's the Detail Tweaker LoRA? And what's the AdverseCleaner Extension?

Anonymous 08/09/2023 (Wed) 00:10:18 No. 5436

>>5432 Based64Mix is a model, basically the AI's "brain", different brains trained in different images give different results, like AbyssOrangeMix, cardos animated or any other model mentioned in this thread. You can find most models at civitai.com, they are labelled as "checkpoints" (another term for the main "brain" of the AI). You download the file, drop it in the following folder: stable-diffusion-webui\models\Stable-diffusion And then select that model from the dropdown at the top left of the UI (takes a while to load). The link to the Based64 model was posted here >>5010 Hed is a ControlNet model, it doesn't generate images directly but it can be used to assist the main model in generating images in a specific way. Search for a ControlNet tutorial, there's a lot to explain to get it working. Adverse Cleaner is an extension to the UI, it's what I use to remove the annoying grainyness that I was having in some of my old pics, for example open the second pic here >>4989 at full resolution, it looks very grainy. You install extensions from the extensions tab (duh) and you just need to copypaste the github url and the UI will download and install the extension on its own, very handy. Adverse Cleaner just adds a new tab to your UI, you go there, drop your grainy pic, fiddle with some bars and it will apply a filter to your pic to make it look smoother.

Anonymous 08/09/2023 (Wed) 00:16:19 No. 5437

>>5432 Oh, I forgot, the detail tweaker is exactly what it says on the tin, it makes images more intricate and complex. You download the add_detail.safetensors file, put it on this folder: stable-diffusion-webui\models\Lora Then you can call the detail tweaker in your prompt to increase or lower detail like this: <lora:add_detail:1> You just put that in your prompt. The number is the amount of emphasis you want to give it. I have calls to this LoRa in my styles.csv that I posted above because I use it a lot.

Anonymous 08/09/2023 (Wed) 20:49:31 No. 5449

lolis IA kh4k4r0t0 lolis 08/11/2023 (Fri) 23:37:19 No. 5471

zombie

Anonymous 08/12/2023 (Sat) 06:45:18 No. 5472

>>5471 Ummm... I'll skip on that one.

Anonymous 08/16/2023 (Wed) 22:04:47 No. 5508

>>5226 >uohtism Excellent name.

Anonymous 08/18/2023 (Fri) 05:29:21 No. 5520

I've been experimenting with automatic1111 webui after getting rocm properly installed to some extent through some kind of divine ritual involving hours of "random bullshit go!" type efforts. I thank the Omnisiah it ended up working in the end Running this on a 8GB VRAM AMD card can be a strenuous effort if you have no idea what you're doing, like me. Here is the command I use to run the web ui: ./webui.sh --no-half --always-batch-cond-uncond --opt-sub-quad-attention --medvram --disable-nan-check --theme dark Trying to push the resolution or batch size too high will predictably result in "out of memory" errors a lot of the time. I tried --lowvram mode, but it slows down the generation speed to a crawl. So I settled with generating batches of four 512x512 images at a time, and then scaling the proper looking ones by 1.6 to a single 819x819 image. I might have the wrong approach to this, but even going into this blind it's been a very fun experience.

Anonymous 08/19/2023 (Sat) 15:00:10 No. 5535

>>5520 To prevent the out of memory. Set the resolution to a typical dimension 4:3 3:2 16:10 using 512 576 640 768 832 960

Anonymous 08/20/2023 (Sun) 17:46:03 No. 5538

flatsims 08/20/2023 (Sun) 20:16:39 No. 5539

yes you always want to start with a common low resolution, 512x512, 512x728, 728x512 and then when you have a decent image, uposcale that to 2x larger or more. Then inpaint if you need too. I do a batch face swap as well as that re-drawing of the face always improves it.

flatsims 08/20/2023 (Sun) 20:18:23 No. 5540

flatsims 08/20/2023 (Sun) 20:21:31 No. 5541

A little retro, soda anyone?

Anonymous 08/22/2023 (Tue) 17:24:22 No. 5564

Anonymous 08/22/2023 (Tue) 19:55:43 No. 5566

Anonymous 08/22/2023 (Tue) 20:18:23 No. 5567

>>5541 That's pretty good.

Anonymous 08/24/2023 (Thu) 05:09:54 No. 5580

Anonymous 08/24/2023 (Thu) 12:38:48 No. 5581

I've been lazy, so here's some new stuff. I have to upload this and the last post to the booru too.

Anonymous 08/24/2023 (Thu) 12:40:44 No. 5582

Just for fun, I grabbed every pic in my img2img folder after finishing the first pic and put them into a sequence. I think it's interesting to see how the image changes as I'm messing with it.

Anonymous 08/25/2023 (Fri) 10:21:11 No. 5597

>>5582 Real fukken nito.

Anonymous 08/26/2023 (Sat) 15:55:20 No. 5609

Can anyone make some lewds for this girl? Hard to say what exactly I want but I left some possible examples. https://civitai.com/models/18345/totally-not-about-ava-and-nothing-of-any-demons-and-such >>5564 >>5566 This stuff's really good

Anonymous 08/27/2023 (Sun) 01:21:52 No. 5615

>>5581 Have you tried experimenting with some meme poses? The AI might be really good at doing things like the jack o pose, or the standing split since those poses should be well tagged.

Anonymous 09/03/2023 (Sun) 14:03:10 No. 5690

Kh4k4r0t0 lolis 09/05/2023 (Tue) 11:22:59 No. 5727

Delicious

Anonymous 09/06/2023 (Wed) 16:51:15 No. 5753

You know that thing artists do where they make variations of the same girl with different (or less) clothes? I figured I'd give that a try, and turns out it's a lot of fun. >>5615 I had already tried the Jack-o pose and couldn't get it right. The tags don't really get you the pose, so I tried manually making an openpose skeleton and using controlnet to generate the pic, but the AI gets extremely confused with that pose and just generates a girl lying on her stomach with legs coming out of her back.

Anonymous 09/06/2023 (Wed) 16:52:26 No. 5754

And a few more pics.

Anonymous 09/06/2023 (Wed) 17:19:29 No. 5756

>>5753 huh, I completely missed the weird color glitch in the microbikini pic. Here it is with another coat of paint.

Kh4k4r0t0 lolis 09/06/2023 (Wed) 20:39:46 No. 5761

Anonymous 09/07/2023 (Thu) 11:28:47 No. 5764

>>5753 Dress up is pretty fun. I'm surprised you got her body to stay consistent with different outfits though. On the sling bikini, it looks like she only has one nipple because one is clearly protruding and the other isn't. I suggest making them both hard. >>5754 Could I request you do Shierke from Berserk. I think her curly round hair would look really good in the styles you use. Is it just me, or is her ass way too low in my first pic? >>5761 Everytime someone posts girls that lean towards realism, they have unpolished hands an pussies. Gee, I wonder why?

Anonymous 09/08/2023 (Fri) 11:19:53 No. 5784

>>5778 >Puts less work into the picture <Calls the more polished AI low effort Lmao.

Kh4k4r0t0 lolis 09/09/2023 (Sat) 18:36:12 No. 5806

Anonymous 09/11/2023 (Mon) 14:59:23 No. 5818

>>5764 Nipple fix and Schierke delivery. >I'm surprised you got her body to stay consistent with different outfits though. It's the same thing artists do, start with the nude body and then paint stuff on top. It just needs the added step of inpainting with AI to blend it in perfectly. >>5806 This is pretty neat. I wonder if you can convert normal pics into statues using controlnet and that model.

Kh4k4r0t0 lolis 09/12/2023 (Tue) 01:26:49 No. 5821

Anonymous 09/12/2023 (Tue) 11:19:32 No. 5823

>>5818 Thanks a ton. You are a saint. >This is pretty neat. You should try some statue girls then. I don't like petrification, and most statue girls I see made by petrificationfags are very low quality, and usually involve the girl dying. But living statue girls are hot.

Kh4k4r0t0 lolis 09/12/2023 (Tue) 13:14:10 No. 5824

Best ai generated loli sites https://allthefallen.moe/forum/index.php?threads/kh4k4r0t0-ai-generated-lolis.51149/ https://booru.allthefallen.moe/posts?tags=user%3Akh4k4r0t0&z=5

Anonymous 09/12/2023 (Tue) 21:44:56 No. 5836

>>5824 You were supposed to use the AI, not become it!

Anonymous 09/13/2023 (Wed) 13:23:09 No. 5841

kh4k4r0t0 lolis 09/13/2023 (Wed) 20:18:40 No. 5853

Anonymous 09/14/2023 (Thu) 22:14:35 No. 5871

Anonymous 09/15/2023 (Fri) 12:44:48 No. 5873

>>5823 >You should try some statue girls then. Turns out converting pics to statues actually works amazingly well. >>5871 How are you getting proper sex scenes? All I ever get with prompting is body horror and disembodied dicks sprouting out of the ground, and that's when I even get something at all, in 90% of cases the AI completely ignores the sex tags and just generates the girl alone.

Anonymous 09/15/2023 (Fri) 21:19:05 No. 5885

Summer night adventure in the woods. (In some alternate universe where mosquitoes won't eat you alive when doing this.)

kh4k4r0t0 lolis 09/16/2023 (Sat) 04:28:49 No. 5889

>>5873 eu queria ter o control net pra poder fazer isso

Anonymous 09/16/2023 (Sat) 14:48:15 No. 5892

>>5873 God damn, that style fits Shierke so fucking well. Thank you anon.

Kh4k4r0t0 lolis 09/19/2023 (Tue) 16:22:15 No. 5916

flatsims 09/19/2023 (Tue) 16:33:39 No. 5917

Proper sex scenes are pretty easy with the right models and sex loras although taking some good scenes and using img2img and possibly controlnet enpose can help too. While you may need to fix a few things in inpainting you can get a lot of quality in any model that isn't a first gen one. Those first ones did suck at that and should be abandoned for newer ones.

flatsims 09/19/2023 (Tue) 16:35:02 No. 5918

Anonymous 09/20/2023 (Wed) 23:50:55 No. 5942

Looking for kinky loli

Anonymous 09/22/2023 (Fri) 00:32:46 No. 5954

>>5917 >>5918 >doesn't say what model it is It's like reading through an old tech support thread.

Anonymous 09/23/2023 (Sat) 23:05:21 No. 5972

>>5853 Nice, but naked would be better. >>5916 >niggers

Anonymous 09/26/2023 (Tue) 05:14:35 No. 5981

>>5972 >naked would be better. Full naked is NEVER better.

Anonymous 09/26/2023 (Tue) 07:29:00 No. 5982

>>5806 >>5873 OOOOH Do some more marble ones. Maybe experiment with other sculpture materials too: bronze, gold, crystal, etc.

Anonymous 09/28/2023 (Thu) 03:08:10 No. 5986

Request: naked loli Shantae surrounded by blue flames.

Teleo 09/28/2023 (Thu) 18:35:11 No. 5987

>>5986 Here you go, and I upscaled it. I asked Stable Diffusion for "1girl" but it generated two Shantaes for some reason, so you get to have a nice front view and back view. To Uohtism-sensei, thanks for all the advice you've given me. I've got more questions: OP mentioned a Krita plugin. What exactly can SD do in Krita? Or what's the advantage of having Krita linked with SD? Also, what does putting numbers in the "Seed" textbox do? I've used that textbox to regenerate old images and to create variations on those old images, but is it useful for anything else? Also, I found out that the "PNG Info" tab on SD can be used to recount the prompts and settings that were used to generate an image. There were two images you made that I said I liked and I asked about the prompts and settings, but you said you forgot the exact prompts and settings. Did you know about the PNG Info tab?

Anonymous 10/01/2023 (Sun) 21:04:04 No. 5991

Between IRL work piling up, a lot of fruitless model hunting and failed experiments while trying to make sex scenes, I'm not feeling very motivated these days, but I'll make the Shantae pic at least >>5986 Model hunting is getting frustrating if you want to make lewd loli, I even upgraded my GPU and tried the SDXL models, but I couldn't get any good results. Seems that SDXL comes lobotomoized out of the box and the models based on it tend to have the same issue. Now the Bing AI came out and it's way ahead of SD in terms of quality outputs, but of course it's also a lobotomized corporate product and you can't even run it locally. In short, all the good new stuff is either censored or gated, and I wasn't able to upgrade my current setup despite a lot of trying, so I'm thinking I'll be going into hibernation until there's some non-lobotomized innovation I can use. >>5987 >OP mentioned a Krita plugin. What exactly can SD do in Krita? I tried that plugin a long time ago and the idea is to use SD directly in Krita instead of using the webUI. That would certainly be very convenient but I ended up not using it because inpainting didn't work very well in that plugin, and now with stuff like ControlNet and the upscaling scripts, I doubt the plugin kept up with all the functionality you would need. >Also, what does putting numbers in the "Seed" textbox do? It's like the seed in any random generator, the idea is to be able to replicate the same result should it be needed. If you use the same prompt, model and seed, you should get the same image every time. >Also, I found out that the "PNG Info" tab on SD can be used to recount the prompts and settings that were used to generate an image. There were two images you made that I said I liked and I asked about the prompts and settings, but you said you forgot the exact prompts and settings. Did you know about the PNG Info tab? I did, but I don't have the original pics anymore, the ones with the metadata saved.

Anonymous 10/03/2023 (Tue) 01:14:37 No. 5994

>>5991 >but I'll make the Shantae pic at least Very nice! Thanks! >so I'm thinking I'll be going into hibernation until there's some non-lobotomized innovation I can use. This is very unfortunate news. I hope something good will turn up soon and you'll be able to unleash your full potential. >I don't have the original pics anymore, the ones with the metadata saved. I don't do anything with AI, but isn't it the best practice to always save metadata?

Anonymous 10/05/2023 (Thu) 22:34:54 No. 6001

>>5972 mad whiteboi spotted

Anonymous 10/06/2023 (Fri) 05:17:35 No. 6002

>>6001 They're not even pretty niggers. They're ugly as fuck.

Anonymous 10/06/2023 (Fri) 11:18:26 No. 6005

>>6001 >>>/delicious/

Anonymous 10/06/2023 (Fri) 22:57:09 No. 6008

>>5991 Something needs to leak already. I feel like things are being held back so much because these corporate faggots are censoring everything.

Sitting on glass Luna 10/08/2023 (Sun) 13:18:02 No. 6018

Sitting on glass

Anonymous 10/10/2023 (Tue) 01:18:55 No. 6027

sea of stars

Anonymous 10/17/2023 (Tue) 16:20:58 No. 6038

Anyone know what happened to Aquapanty on Pixiv ?

Anonymous 10/27/2023 (Fri) 15:20:02 No. 6071

I'm back. There are no new leaks so I just went back to doing my same old thing.

Anonymous 10/28/2023 (Sat) 02:43:09 No. 6072

>>6001 no-brain nigger monkey spotted

Anonymous 10/28/2023 (Sat) 15:51:42 No. 6073

>>6071 Glad to see you're alive. Sad to see proprietary shit being ass as usual. Would it be possible to make a loli Demencia?

Teleo 10/28/2023 (Sat) 22:26:41 No. 6074

>>6071 I made these images with you in mind. Thank you Uohtism-sensei for all the guidance and advice for SD you've given, and for all the great images you've shared. By the way, I've started using Adverse Cleaner. Sometimes when I generate an image using AOM3 or B64M, I'll get white and purple splotches on the skin. Adverse Cleaner is good at eliminating white splotches, like for the last two images I've posted here, but I still regularly get purple splotches. Any advice? I've just been using the default settings in Adverse Cleaner, but what do your settings in Adverse Cleaner normally look like? Happy Halloween!

Anonymous 10/29/2023 (Sun) 07:28:34 No. 6076

>>6071 Good to have you back.

Anonymous 10/31/2023 (Tue) 10:29:59 No. 6079

Does someone still have the squeezer lora?

Teleo 11/01/2023 (Wed) 07:21:19 No. 6081

>>6079 This anon posted the link: >>4947

Anonymous 11/02/2023 (Thu) 22:00:31 No. 6082

Anonymous 11/02/2023 (Thu) 22:01:22 No. 6083

Anonymous 11/03/2023 (Fri) 13:25:35 No. 6084

>>6082 >>6083

Anonymous 11/03/2023 (Fri) 22:02:11 No. 6085

>>4000 Is there any downloads or all in one packs that you guys would recommend? I only started messing around with automatic1111 recently and I just want to make some cute lolis and hebes

Anonymous 11/04/2023 (Sat) 13:01:17 No. 6086

>>6085 >hebes You best avoid that pedospeak, outlander. Folks don't take kindly to your sort around these parts.

Anonymous 11/04/2023 (Sat) 20:37:45 No. 6089

>>6073 That's a very tricky character design, this is the best I could do. I might give it another shot some other day. >Sad to see proprietary shit being ass as usual. It's downright hilarious how much they're lobotomizing DALL-E 3, even normies are getting annoyed. >>6074 Thanks anon, I appreciate it. One more tip: I noticed some of your gens look grayish and washed out, that usually happens when you're not using a VAE, or using the wrong VAE. For me, I just found ClearVAE a while back and it's been working consistently with my models so I haven't bothered to change it. To use ClearVAE, just download it from civitai: https://civitai.com/models/22354/clearvae Drop that file in your models\VAE folder, then go to Settings -> VAE, select the file from the dropdown and click "Apply settings" at the top. The second pic here is a comparison of what it does.

Teleo 11/04/2023 (Sat) 23:00:42 No. 6090

>>6089 Thanks, I was wondering why my generations looked so washed out. Anon delivers again. But what about Adverse Cleaner and the white and purple splotches I'm getting? Would ClearVAE help with that?

Anonymous 11/04/2023 (Sat) 23:20:39 No. 6091

>>6090 The VAE should solve purple splotches, if you look in the comparison pic, you can see a purple spot on the girl's shoulder in the pic without the VAE. That was supposed to be a shiny spot but without the VAE the AI derped out and made it purple. As for my Adverse Cleaner settings, I posted them here >>5148 You can put the following in your ui-config.json to get my settings as default: "adverse_cleaner_tab/Diameter (default = 5)/visible": true, "adverse_cleaner_tab/Diameter (default = 5)/value": 1, "adverse_cleaner_tab/Diameter (default = 5)/minimum": 1, "adverse_cleaner_tab/Diameter (default = 5)/maximum": 30, "adverse_cleaner_tab/Diameter (default = 5)/step": 1, "adverse_cleaner_tab/SigmaColor (default = 8)/visible": true, "adverse_cleaner_tab/SigmaColor (default = 8)/value": 4, "adverse_cleaner_tab/SigmaColor (default = 8)/minimum": 1, "adverse_cleaner_tab/SigmaColor (default = 8)/maximum": 30, "adverse_cleaner_tab/SigmaColor (default = 8)/step": 1, "adverse_cleaner_tab/SigmaSpace (default = 8)/visible": true, "adverse_cleaner_tab/SigmaSpace (default = 8)/value": 4, "adverse_cleaner_tab/SigmaSpace (default = 8)/minimum": 1, "adverse_cleaner_tab/SigmaSpace (default = 8)/maximum": 30, "adverse_cleaner_tab/SigmaSpace (default = 8)/step": 1, "adverse_cleaner_tab/Radius (default = 4)/visible": true, "adverse_cleaner_tab/Radius (default = 4)/value": 1, "adverse_cleaner_tab/Radius (default = 4)/minimum": 1, "adverse_cleaner_tab/Radius (default = 4)/maximum": 30, "adverse_cleaner_tab/Radius (default = 4)/step": 1, "adverse_cleaner_tab/Accuracy (default = 16)/visible": true, "adverse_cleaner_tab/Accuracy (default = 16)/value": 8, "adverse_cleaner_tab/Accuracy (default = 16)/minimum": 1, "adverse_cleaner_tab/Accuracy (default = 16)/maximum": 30, "adverse_cleaner_tab/Accuracy (default = 16)/step": 1,

Anonymous 11/05/2023 (Sun) 01:06:03 No. 6092

>>6086 ? That's been a phrase since the early 2000s. I guess you guys don't categorize cute drawn girls here.

Anonymous 11/06/2023 (Mon) 11:12:27 No. 6094

>>6092 Nobody said it was a new term. All across the net, practically no one uses that term with regard to 2d, and the /hebe/ board that once existed was exclusively about discussing 3DPD, until it was banned, because pedoniggers can't not post illegal material.

Anonymous 11/06/2023 (Mon) 11:16:38 No. 6095

>>6089 >That's a very tricky character design, this is the best I could do. That's pretty damn fucking fantastic. >It's downright hilarious how much they're lobotomizing DALL-E 3, even normies are getting annoyed. It's AIDungeon all over again. Except this time there's competing alternatives even before the leading product goes full retard.

Anonymous 11/09/2023 (Thu) 19:36:55 No. 6103

>>6095 Thanks anon, I made another one, trying to get a few more details accurate to the design. Also new loli Ryuuko.

Anonymous 11/10/2023 (Fri) 07:06:00 No. 6104

>>6100 You're not welcome here. Leave.

Anonymous 11/10/2023 (Fri) 07:07:51 No. 6105

>>6103 Nice.

Anonymous 11/10/2023 (Fri) 15:30:29 No. 6106

>>6103 These two both look really good, but I think Demencia's pussy might be a bit lopsided, and Ryuko's right nipple viewer's left is a little too undefined compared to the other.

Teleo 11/10/2023 (Fri) 20:03:31 No. 6107

>>6091 Thanks, Anon. That VAE helped with all my generations using AO3 and B64M. I was wondering what was the point of those two checkpoints since all their generations had dull colors, but I guess that was a problem on my end. And your Adverse Cleaner settings also helped a lot. Here's something I've found out: If you use "reference_sheet" as a tag, SD will create an image of multiple girls with the same or similar appearances from more than one angle, that's how I got that nice butt shot in the fourth pic here. If you try it and it doesn't work, just try increasing the tag strength.

Anonymous 11/12/2023 (Sun) 19:01:54 No. 6112

>>6106 >I think Demencia's pussy might be a bit lopsided, and Ryuko's right nipple viewer's left is a little too undefined compared to the other. hmmm, yeah I can see it, here are some fixes. >>6107 >Here's something I've found out: If you use "reference_sheet" as a tag, SD will create an image of multiple girls with the same or similar appearances from more than one angle, that's how I got that nice butt shot in the fourth pic here. That's fucking genius. In a pic with multiple characters the AI will just repeat the same character most of the time, that tag is a really clever way to turn that problem into an advantage.

Anonymous 11/13/2023 (Mon) 02:34:21 No. 6113

>>6112 Can you make more like the fourth pic? She's beautiful.

Anonymous 11/13/2023 (Mon) 11:33:30 No. 6114

>>6112 Danke. You're a saint.

Anonymous 11/14/2023 (Tue) 09:11:59 No. 6115

bericbone stopped uploading to sadpanda, does anyone know where to get his latest?

Anonymous 11/18/2023 (Sat) 03:57:21 No. 6132

>>6115 whats sadpanda?

Anonymous 11/18/2023 (Sat) 09:02:43 No. 6133

Does anyone have more like this?

Anonymous 11/18/2023 (Sat) 16:35:23 No. 6136

>>6132 WELCOME NEWFRIEND!

Anonymous 11/18/2023 (Sat) 20:19:37 No. 6138

>>6136 You don't need to be a cunt, anon. Just answer their question.

Anonymous 11/19/2023 (Sun) 05:00:00 No. 6140

>>6138 Rule #1 of sadpanda, we do not talk about sadpanda. Rule #2 of sadpanda, we do not talk about sadpanda. exhentai.org You used to be treated to a picture of a sad panda when you went here, unless you sucked over 9000 cocks and had access.

Anonymous 11/19/2023 (Sun) 06:48:28 No. 6141

>>6140 zomg whats the password i cant get in

Anonymous 11/19/2023 (Sun) 16:23:57 No. 6142

>>6138 >You don't need to be a cunt, anon. Just answer their question. If you think that is being a cunt, maybe imageboards aren't for you? Are you even old enough to post here?

Anonymous 11/19/2023 (Sun) 20:27:15 No. 6144

>>6132 ligma balls

Anonymous 11/19/2023 (Sun) 21:03:48 No. 6145

>>6138 Yeah, I gotta agree with >>6142, that isn't even remotely being a cunt, and I've been treated far worse on this site alone. I would say grow thicker skin, but he didn't actually say anything bad to you. Either learn not to be a retard or stick to forums. >Are you even old enough to post here? I doubt he is. I personally don't really about his age, I care if he's acting like a retard or not. >>6133 You can find more on Sankaku. >>6144 Insert the correct CDs.

Anonymous 11/19/2023 (Sun) 22:05:02 No. 6146

Two new pics. I'll have to get in the habit of working on a pic for at least an hour per day because doing it all on weekends takes up a significant chunk of the day.

Bericbone Anon57 11/20/2023 (Mon) 12:53:27 No. 6147

What happened to AIunlimited section group? Where can we find his works?

Anonymous 11/20/2023 (Mon) 17:53:18 No. 6148

>>6145 > I personally don't really about his age, I care if he's acting like a retard or not. 4chan was created by an underage retard, but both were still bullied. Bullying is a way of life.

Anonymous 11/21/2023 (Tue) 01:45:18 No. 6149

>>6141 What part of sucking over 9000 cocks didn't you understand? A quick search of the forums will reveal that no one understands what the rules are for getting in. Brony gives and brony takes away.

Lolis Gumtree917 11/25/2023 (Sat) 16:04:20 No. 6163

Hi where can i find the Loli images created by others? Are there any session or other groups i can join? AI Unlimited on session is not working anymore. Thanks!

Anonymous 11/28/2023 (Tue) 17:31:55 No. 6169

How to prompt NSFW???

Anonymous 11/28/2023 (Tue) 20:37:56 No. 6170

I would really like to know how to add NSFW on Stable Diff. I can get stuff like this, but it's missing, you know, that.

f u 11/29/2023 (Wed) 05:31:19 No. 6175

>>6170 >>6169 Get yer 3d ai bullshit outta here

Anonymous 11/29/2023 (Wed) 12:48:15 No. 6178

>>6175 Hey, his first one isn't bad. Those are big attractive disney eyes.

Teleo 11/29/2023 (Wed) 14:03:08 No. 6179

I love a loli (lacking) in armor. >>6112 >That's fucking genius. In a pic with multiple characters the AI will just repeat the same character most of the time, that tag is a really clever way to turn that problem into an advantage. Yeah, I wanted front-and-back sort of pictures like your fourth pic, and I saw somewhere in this thread someone said these checkpoints were trained on booru images and tags. I've seen reference sheet pics on Danbooru before, so I thought that was probably the best way to achieve that kind of picture. >>6169 >>6170 Did you type “NSFW” in the prompt? Try using a different checkpoint. Not all checkpoints are made for NSFW pics. You can find other checkpoints on civitai.com and huggingface.co. The download page for a checkpoint will usually tell you if it's NSFW-friendly or not.

Anonymous 11/29/2023 (Wed) 20:51:47 No. 6180

>>6175 Sorry my guy, didn't meant to be a prick. Just experimenting stuff

Anonymous 12/07/2023 (Thu) 13:37:30 No. 6192

https://www.pixiv.net/en/artworks/114031534

Anonymous 12/20/2023 (Wed) 19:19:47 No. 6231

Have some Christmas themed lolis. I finally found out what was making my pics so grainy. Turns out the ESRGAN 4x upscaler was adding a lot of noise to the imag. Using the ultimate SD upscale script with Lanczos, the Euler A sampler and around 0.2-0.25 denoising strength produces a much cleaner image. With this method I don't even need Adverse Cleaner most of the time.

Anonymous 12/24/2023 (Sun) 17:09:00 No. 6238

I've been doing some experiments with clothing layers. It takes some work, but I like the way you can gradually undress the loli. Sadly I don't know much about fashion, so I'll take suggestions about to what type of clothing to add.

Anonymous 12/24/2023 (Sun) 17:10:51 No. 6239

And here is one with 4 layers. In other news, looks like someone created a tag for me in atfbooru, now instead of looking at my user you can search the tag "uohtism_(ai_generator)". Anyway, merry Christmas, anons.

Anonymous 12/26/2023 (Tue) 06:27:38 No. 6243

>>6239 I think the prompt for flat chest creates an unnatural looking concave ribcage. There has to be a better prompt.

Anonymous 12/26/2023 (Tue) 16:00:14 No. 6244

>>6243 >unnatural looking concave ribcage I don't see it.

Anonymous 12/27/2023 (Wed) 12:53:00 No. 6251

If I may make a request, I would like to request some very hardcore Flande Scarlet + Remilia Scarlet Yuri (fisting, anilingus, 69 cunnilingus, watersports, licking cum from pussies/assholes, orgies, etc.), or advice for how to make good versions of it myself using PixAI or NovelAI. Here's what little I was able to do on my own, but despite these few successes most of my results have been failures. And yes, I am using negative prompts, so I figure the issues must be not knowing what loras or models to focus on for what I seek to create.

Anonymous 12/27/2023 (Wed) 21:12:18 No. 6255

>>6244 see the 2nd pic in >>6239 and both ribbon pictures in >>6231 as most recent examples. and even the last one. It's more apparent in clothed pictures, but some pictures earlier in this thread are like that too. In many of them the ribcage actually runs concave if you follow the lines or perfectly flat rather than the natural curve of a ribcage. This makes the final result uncanny. You can see may of the girls in >>6107 have ribs (starting from the diaphragm) start narrow, then get wider, then suddenly dip inward at the chest. In the 4th image the girl on the left has a clear indent at the chest if you look under the arms. In pictures related look at the right side of the image (her left) of the fairy. How the chest dips inwards. or the right side of the amazon how the lower ribs bulge out compared to where the chest is. In contrast to the rainbow haired girl who's chest came out somewhat natural looking by comparison even though you can tell the AI attempted to flatten the chest. I think what's happening is its giving them small breast meat, then reducing the curve of the ribcage to make the final result "flat" rather than just make less breast meat on the ribcage. yes i am aware this is autism

Anonymous 12/27/2023 (Wed) 23:24:59 No. 6257

>>6255 I'm trying, but I don't see it. The chest curve looks fine to me. I can see the weird bulge on the ribs in the loli amazon pic you posted, but if you trace a curved line through that bulge, the chest itself is in the right place. Keep in mind that all of the pics you're pointing to have the girls arching their back, and that last pic of the rainbow haired girl you posted as a good example has the back arched forward instead of backward.

Anonymous 12/28/2023 (Thu) 00:10:44 No. 6258

>>6257 I didn't use your example as an example for a reason, it's harder to see because of how the boob is drawn, but the clothes make it obvious. On her left breast (right side) the breast meat is level with the ribcage below it. which would mean the ribcage is below that. one boob depth below. Try tracing the "boob" on its own, then draw where you think the ribcage is under it. The boob should be outside of that circle, not in-line with it as it is in your example.

Anonymous 12/28/2023 (Thu) 00:40:17 No. 6259

>>6257 >>6258 I tried my best here to illustrate what i'm talking about. the breast meat should be on top of the ribs. not in-line. when they are in-line with the ribcage, it implies the ribs are below it, and therefore concave to fit the breast tissue, which the AI draws into the shape of the underlying ribs like breasts drawn onto a lego figure. flat chests arent supposed to be actually flat. one of these images is a male and many of the chests in the thread follow the red line, which is even flatter than a male chest. when it should be atleast green (meat on top of bone instead of skin stretched over bone, but drawn as if its a meaty soft chest)

Anonymous 12/28/2023 (Thu) 01:41:46 No. 6260

>>6258 >>6259 So, something like this? Ehhhh I'm not convinced. This doesn't look right to me. Her chest is too small and at a too forward facing angle to be protruding to the side like this. This just makes it look like one of her breasts has a tumor or something. I'd just have to make both breasts bigger to make this look consistent, and at that point I'd just be giving the girl bigger breasts, so it's not a matter of perspective anymore.

Anonymous 12/28/2023 (Thu) 02:31:21 No. 6264

>>6260 Kinda. it's going to look wrong as long as the AI prompt is wrong because the problem starts with the anatomy and cant be fixed by just adding a small amount of breast meat to one side of the nipple. the ribcage is still wrong and the nipple is out of place. But yes, breast meat goes on top of the bone. there isnt a cavity in peoples rib cages for the breast to form inside of that makes it flush/level with the ribs. the breast should always sit on top and protrude outward instead of being perfeclty level with the ribs below. im almost positive this deformity is from literally entering in "flat chest" because characters with breasts and even oppai loli dont have this problem. and AI doesnt know what a flat chest is other than a literally flat chest, which does not exist in humans. It's hard to find AI examples that are correct because its screwed up by AI doing small breasts even more frequently than it fucks hands up. Heres 2 angels. one with the problem, one without. and heres a fairy without the problem, even though its covered with a bra. and 2 from the same set, one with the problem and one without but one the breasts are too large but atleast her ribs arent deformed.

Anonymous 12/28/2023 (Thu) 02:41:42 No. 6265

>>6260 Using your example. if you use the breasts as a guide for where the ribs are (slightly below the breast) the ribs on that image follow the red line, which as you can see curves inward. concave chest. the breast/puffy nipple pokes out (green line) and only comes level with the rest of the ribs. which is wrong, it should protrude. when AI combine it, its a solid line (the oval you drew earlier), but shouldn't be as breasts are on top of the ribs.

Anonymous 12/28/2023 (Thu) 10:54:02 No. 6269

>>6264 >>6265 No need to keep explaining it, it's already clear what you mean, it's just not there. In your second post, I don't see any reason to curve the red line inward like that, at that angle, there's not enough breast flesh to need a cavity like that. The examples you posted as not having the problem have just been girls with larger, more developed breasts. I think you're interpreting the breasts in flat chested girls as larger than they actually are.

Anonymous 12/28/2023 (Thu) 22:59:17 No. 6273

>>6269 >I don't see any reason to curve the red line inward like that The red line is just following the line of the ribcage. It curves inward because the breast has mass. You can gauge the mass with your eyes, but if you can't see it you can try drawing contour lines from the bottom of the breast up through the nipple, and where the center of breast ends. She is drawn flatter than an underweight boy, while the texture of the breast is puffy creating an uncanny chest area. If you can't see it, i guess i can't help you. Last example, but this one is very guilty of it. even you should be able to spot where the ribcage dips inward

Anonymous 12/29/2023 (Fri) 01:58:42 No. 6275

>>6238 >>6239 >so I'll take suggestions about to what type of clothing to add. Loose fitting clothing where it appear like the girl might be oblivious to the bits she's showing off is almost always a win in my book. Merry Christmas and Happy New Year. >More loli Jessica Fuck yes.

ayaa 01/02/2024 (Tue) 06:01:40 No. 6280

ya

Teleo 01/03/2024 (Wed) 18:38:10 No. 6281

>>6231 Belated merry Christmas and happy near year, Uohtism-sensei. Wow, those images are fantastic! I don't see how artists could get mad at AI images, the way you work on your images is pioneering and it's probably how most artists will make their stuff in the future; i.e. using AI for broad strokes and then adding the human touch to flaws and small details. That's how people have been using Photoshop to alter photography ever since that application existed. By the way, what's the "ultimate SD upscale script"? Are you referring to the SD upscale script already in SD, or is this ultimate version an add-on or update? Also, is it just me, or is it getting harder to make lolis without loras in SD 1.5/1.6? It used to be really easy, but now it seems the whole program tries to make post-pubescent characters in place of lolis. I first noticed this phenomenon with the Cardos checkpoint, but even other checkpoints I have are doing this. I've gotten so paranoid about my SD 1.6 going anti-loli that I haven't updated it since mid-December (except for my extensions). Am I going overboard? Am I imagining things? Are the updates worthwhile enough that I should just bite the bullet and keep updating it and just depend on the loras?

Anonymous 01/05/2024 (Fri) 13:43:37 No. 6283

>>6281 What application are you using to generate these ai images? The freemimum ones that I have been using went anti-NSFW recently.

Teleo 01/05/2024 (Fri) 16:16:00 No. 6284

>>6283 See >>4000

Anonymous 01/07/2024 (Sun) 11:57:50 No. 6289

>>6281 >>5917

Teleo 01/07/2024 (Sun) 22:29:09 No. 6295

>>6289 That doesn't really answer my questions. Take Cardos for example; it used to be I would simply put "loli" and "NSFW" in the prompt field with no loras and it would give me an NSFW loli, plain and simple, but now it gives me a postpubescent character. I like loras, but I think depending on them too much provides samey results; I got more variety when I just let the model have its freedom. And this was early on before I started regularly applying all the SD 1.5 updates from Git, which makes me think anti-loli bias has been implemented in the application. I opted out of automatic updates when I first installed SD, instead I enter "git pull" in the command prompt for them; I'm sure that updates SD but I'm not so sure it updates the models, but I swear I had this problem with Cardos before it happened with any other model.

Anonymous 01/08/2024 (Mon) 02:16:34 No. 6297

>>6295 Try these https://huggingface.co/JosefJilek/loliDiffusion

Anonymous 01/10/2024 (Wed) 21:44:33 No. 6301

>>6297 Not him, but for future reference once I have a setup worthy of AI, how would I install this? Not seeing a download button anywhere.

isabelle 01/11/2024 (Thu) 03:29:00 No. 6302

creat isabelle from animail crossiing when she is 13 on her birthday and is being raped by her twin brother digby

Teleo 01/11/2024 (Thu) 06:41:51 No. 6304

>>6297 I downloaded that whole package. I think these overall improved my loli generations, but they didn't really solve the systemic issues I think I'm facing, which is fine anyway because there are easy ways around it such as loras and other things at the current moment. But still, thank you. >>6301 Assuming you followed all the instructions in OP's link to get Stable Diffusion on your machine, click the ellipses near the top-right of the huggingface.co page and then click "Clone repository". The window that pops up will tell you you'll need Git LFS, so just go ahead and download that from git-lfs.com. Once you've done that, click "copy" in the window, which will copy the necessary command line needed for downloading (stick with HTTPS and don't bother clicking SSH, in case you were curious about that). Open the command prompt and paste the command line from your clipboard. By the way, once you do get around to downloading Stable Diffusion, SD 1.5 is what you want; I haven't seen any AI erotic artist take 2.0 or 2.1 seriously. Anyway, here are some Italian Renaissance lolis. I'd use Adverse Cleaner to clear up the graininess but I'm currently having an issue with it. Also, I'm so fed up with fucking around with hands. I might just give all my lolis gloves from now on because gloves don't look nearly as bad as deformed hands. At least with a deformed glove, one can assume an anatomically correct hand is inside.

Teleo 01/11/2024 (Thu) 15:45:46 No. 6307

>>6273 Hey, Uhohtism-sensei, have you encountered this problem with Adverse Cleaner? I followed the instructions it gives, but nothing happens.

Anonymous 01/12/2024 (Fri) 10:25:08 No. 6309

>>6238 >>6239 I am VERY interested in how you achieve this? I'm still very new to all of this (new PC got set up yesterday) so most of my gens look like shit right now, but once I get better I would love to be able to do exactly this.

Anonymous 01/16/2024 (Tue) 02:46:49 No. 6319

So, is there hope that someone can make SDXL not suck?

Anonymous 01/17/2024 (Wed) 14:23:39 No. 6321

>>6169 >>6170 I might be able to get around this. What's the checkpoint, AniMerge?

Anonymous 01/19/2024 (Fri) 13:33:18 No. 6327

AI generation attempt with Pokemon best girl.

Anonymous 01/20/2024 (Sat) 02:20:50 No. 6328

TBH, I'm an absolutely huge footfag...what's the state of AI-gen feet in 2024? Faces, bodies, all can be cute, but if feet are still a horror-show on everything that isn't Bing, I might stay away for a little bit...

Anonymous 01/21/2024 (Sun) 03:52:28 No. 6330

>>6307 I had that problem a while ago in my old computer. I don't remember exactly how I solved it, but basically what I had to do was copy a folder from the python install directory into the Automatic1111 UI folder. The command prompt window should tell you the path to the folder it's looking for. >>6309 Basically you start by generating a nude girl, then grab the generated image, upscale it, polish it, (fix errors, add detail to the eyes, etc.) and once the nude pic is perfect, open the pic in any image editing software (I use Krita) and manually paint a clothing item on top, then load the modified pic in img2img with the same prompt as the base image, but adding tags for the clothing item you painted, and use inpainting to blend it in perfectly. Here is a perfect example of how to add stuff manually to an image >>5153 And here >>5389 is an important tip for inpainting at high resolutions. Anyway, I need to get back to making pics. All of my free time's been devoured by Bannerlord and Kenshi over the past month, those games are crack.

Anonymous 01/21/2024 (Sun) 04:02:45 No. 6331

>>6328 Everything can be fixed with inpaitning to get nice feet. See the second pic here >>5581

Anonymous 01/21/2024 (Sun) 20:07:07 No. 6332

Did you make this uohtism, and if so, is it a particular character? It was in my de-aged folder, but I think that might have been an accident.

Anonymous 01/21/2024 (Sun) 20:31:08 No. 6333

>>6332 Yeah, the pic is posted in this thread. She's not a particular character, if she looks similar to anyone it's a coincidence.

Anonymous 01/21/2024 (Sun) 23:36:51 No. 6334

>>6333 Danke.

Anonymous 02/16/2024 (Fri) 06:46:40 No. 6404

>>6335 >>6397 >>6398 If you wanna bump, then just bump.

Anonymous 02/27/2024 (Tue) 02:38:31 No. 6416

A bit from NAI v3

Anonymous 02/27/2024 (Tue) 02:43:19 No. 6417

Anonymous 03/21/2024 (Thu) 13:10:32 No. 6455

Sorry about that whole being dead business. I'm still using my old tools but I've learned a new trick. As >>6304 mentioned, a lot of models are not trained for loli at all, so you tend to get these slightly weird lolis with long limbs and stretched torsos. The loliDifussion models posted here >>6297 do generate lolis with more compact, cuter proportions, but they also struggle with the style tags I like to use and generate more anime-like pics with flatter shading. To me that's kind of a step back because what I like about AI is doing stuff that would be very difficult to do by hand, that's why I like the more painterly and detailed styles. Anyway, the solution turned out to be very simple: first generate a base pic with loliDiffusion at low resolution (the model understands proportions better at low resolution, por example I use 640x768 for the base pic). The loliDiffusion_Based67 variant gives me the cuter proportions, so I start with that, then I go to img2img, switch the model to Based64mix-V3, add in my style tags and regenerate the pic at 0.5 denoising strength. Doing this effectively keeps the body proportions and overall composition from loliDiffusion but changes the art style to be far more detailed and more painterly, just the way I like it.

Anonymous 03/21/2024 (Thu) 13:13:25 No. 6456

>>6455 For comparison, the base loliDiffusion pics before applying the transformation look like this.

nude chubby pussy lotta plump 03/21/2024 (Thu) 15:34:15 No. 6457

blonde hair red bow cute

nude chubby pussy lotta plump 03/21/2024 (Thu) 15:38:00 No. 6458

blonde hair red bow

Anonymous 03/21/2024 (Thu) 16:38:08 No. 6459

>>6457 >>6458 Ok, here's a quick one.

Teleo 03/26/2024 (Tue) 18:05:00 No. 6466

>>6455 Hello, Uohtism-sensei. I think I figured out why I noticed what I perceived to be "anti-loli bias" in my generations. I used to copy your prompts and negative prompts word-for-word, including the loras, and "bad_prompt_version2" in your negative prompts. I eventually did get all the loras you use, and I did get the "bad_prompt" extension, without changing how I prompted. Then I noticed my lolis started getting torsos that were very muscley, detailed, and unloli-like. I recently removed "bad_prompt_version2" from my negative prompts and my lolis' bodies looked way more Democrat supporting. So maybe that was the problem I was experiencing? I know you use "bad_prompt" too, but I don't think it's much of a problem for you since you prefer to make lolis that look a lot more abstract. Anyway, here are more renaissance lolis, the left pic is from my usual process and the right was me attempting the process you described. What do you think? By the way, I never managed to fix my Adverse Cleaner extension; I didn't find that folder you were talking about. Also, you never explained the ultimate SD upscale script.

Teleo 03/26/2024 (Tue) 18:12:22 No. 6467

>>6466 >>6455 A certain word I typed in my previous post was automatically changed to "Democrat supporting". IDK what that's about. What I originally wrote was the word opposite of "postpubescent".

Anonymous 03/26/2024 (Tue) 18:16:05 No. 6468

>>6467 Global word filters put in place to prevent pedos spamming certain words used by FBI run bots that trawl for certain keywords to find and shut down pedo sites. There was an attack on the site 2 years back to flood boards with these key words to shutter the site.

Teleo 03/26/2024 (Tue) 18:20:47 No. 6469

>>6468 Ah, I see.

Anonymous 03/27/2024 (Wed) 14:32:27 No. 6471

>>6466 >I recently removed "bad_prompt_version2" from my negative prompts and my lolis' bodies looked way more Democrat supporting That's weird, bad_prompt and easynegative essentially do the same thing, the only difference is that easynegative does it with a LoRa while bad_prompt uses textual inversion, so I guess bad_prompt is a bit redundant, I just kept it there out of habit really. >the left pic is from my usual process and the right was me attempting the process you described. I like the left better actually, the right pic has cuter proportions but the transformation made the eyes and nose too realistic, which look a little freaky with anime proportions. When that happens I usually do a second inpaint of the eyes and nose with a more anime-like art style. >Also, you never explained the ultimate SD upscale script. Oh, sorry, for some reason I thought I had explained it, but it's not in the thread. Anyway, Ultimate SD Upscale is an extension, you can install it like any other extension using the github URL: https://github.com/Coyote-A/ultimate-upscale-for-automatic1111 Remember that scripts dropdown where you have the "SD upscale" option to increase the resolution of your pics? Well, this extension will install a new option in that same dropdown called "Ultimate SD upscale", which is basically a better upscaling script. The first pic are my current settings for upscaling with that script (you do upscaling from the img2img tab, by the way). If you use those settings you shouldn't even need Adverse Cleaner anymore because that generates a smoother upscale without all the graininess of my old methods. Also, here are some more lolis.

Anonymous 03/28/2024 (Thu) 17:19:00 No. 6473

Any idea on how to fine tune a model using a particular image set?

Teleo 04/01/2024 (Mon) 22:40:00 No. 6493

>>6471 Are those really the settings you use for ultimate SD upscale? You don't use any of the sub upscalers? You keep the sampling steps at 28? You don't apply any kind of seam fixing? I tried upscaling an image by 3 like what you have here, with all these same settings, and the seams were very clearly visible. Pics unrelated; they were upscaled the old-fashioned way.

Anonymous 04/02/2024 (Tue) 21:47:04 No. 6502

(1.29 KB 70x56 Screenshot 2024-04-02 184229.png)

>>6493 Yep, pretty much. Most of the time, Seams fix makes no difference for me, and sometimes gives me worse results, so I don't bother with it. Same thing with the upscalers except for ESRGAN which makes the image noticeably grainier and it's the whole reason why I had to use Adverse Cleaner back when I was upscaling with ESRGAN. Here is a very quick example with zero retouching. Pic 1 is the original. Pic 2 is the upscale with the setting I posted. Pic 3 is with Seams fix. Pic 4 is with Seams fix and ESRGAN. Seams fix did literally nothing, and ESRGAN only made the image grainier. I suppose it bears to mention that I haven't update my Automatic1111 UI in a while. Checking the repository, I have over a thousand commits to pull. Looking at the log, I haven't updated since August 31st, so maybe that's making a difference? These pics should still have the embedded data in case you want to try to reproduce the results.

Anonymous 04/02/2024 (Tue) 21:54:29 No. 6503

>>6493 >>6502 Welp, looks like 8ch automatically deletes embedded data from pics, so I uploaded them somewhere else: Pic 1: https://files.catbox.moe/487t07.png Pic 2: https://files.catbox.moe/ibtc2w.png Pic 3: https://files.catbox.moe/x791d6.png Pic 4: https://files.catbox.moe/e4c8ed.png

Teleo 04/03/2024 (Wed) 07:47:33 No. 6505

>>6502 >>6503 Alright, so I upscaled 478t07.png and ended up with pretty much the exact same results as you; not a seam in sight. And you were right about ESRGAN, it was entirely responsible for the graininess. I typically use a lot of tags and LORAs when generating an image, so I think probably one or more of the LORAs I normally use is responsible for the noticeable seams. Anyway, thanks for all the help, you're awesome! And hopefully my own experimentation gives you some insight. I've been keeping up with SD's updates recently and I don't think you're missing anything.

Teleo 04/03/2024 (Wed) 15:33:50 No. 6506

>>6502 >>6503 >>6505 Your generations definitely have brighter colors than mine. Do you know why that is?

Anonymous 04/03/2024 (Wed) 21:36:27 No. 6509

>>6506 Oh, that's because I forgot to reenable color correction in the im2img settings. I sometimes disable that when changing art styles to avoid some weird color artifacts, but since the option is hidden away in the settings, sometimes I forget to reenable it, which tends to slightly increase saturation each time you run the pic through img2img. The colors in your upscales are actually more accurate to the original, mine got oversaturated because of that option I forgot to check.

Teleo 04/04/2024 (Thu) 09:13:44 No. 6511

>>6509 Oh, I see.

Anonymous 04/07/2024 (Sun) 03:29:11 No. 6515

Old gens but never had the opportunity to share them before

Anonymous 04/08/2024 (Mon) 13:53:52 No. 6516

Nsfw loli

Anonymous 04/08/2024 (Mon) 18:13:23 No. 6517

Some new pics. In case you're wondering about the last pic, I used this: https://huggingface.co/monster-labs/control_v1p_sd15_qrcode_monster

Anonymous 04/10/2024 (Wed) 16:42:52 No. 6518

I wasn't entirely happy with the clothed versions of the elf loli, so I gave it another shot.

Anonymous 04/25/2024 (Thu) 22:19:58 No. 6547

>>6455 Glad to see you're still hammering away at it and refining your AI tard wrangling. Can you make a nopan variant of that second suspenders loli? Also, that Lina Inverse loli might have something weird going on with her ass. Her right (viewer's left) ass cheek that's visible from the front looks like it's going inside of her pussy, sort of hanging out of it. >>6466 >Democrat supporting Lmao, wordfilters.

Anonymous 04/27/2024 (Sat) 14:24:46 No. 6553

>>6547 Have a variant, a fix and a new pic.

Anonymous 05/02/2024 (Thu) 13:40:44 No. 6568

Anonymous 05/02/2024 (Thu) 20:15:50 No. 6572

>>6553 Saintly.

Anonymous 05/04/2024 (Sat) 11:45:44 No. 6573

Anonymous 05/09/2024 (Thu) 16:53:08 No. 6577

>>6328 Good in XL models but going beyond 1024x1024 ratios can introduce extra toes.

Anonymous 07/30/2024 (Tue) 21:40:04 No. 6709

Yo, Uotism? You alive? It's been over 3 months, and I think that's a new record.

Anonymous 08/03/2024 (Sat) 14:08:07 No. 6730

>>6709 Sorry about that, I'm alive, just very lazy and unmotivated. The new XL models have been giving me a hard time because of the censorship that resulted from the anti-AI push. Now that loliDiffusion XL finally released, one of the censorship issues has been solved, but my biggest problem now is that all of the art style tags that I've built over time pretty much stopped working completely in the XL models because people don't train models on artist names anymore. I can generate generic anime loli just fine, but I feel like that's kinda missing the point of using AI. Like I said before, the thing I like about AI is doing stuff that would be extremely difficult or impossible to do by hand, so if I can't get the art styles that I like, I don't feel like I have much to offer. Also, a new wave of people have started using AI to make lewd art and I feel like I have fallen behind by quite a lot. The solution I've been marinating in my head was to use the art style switch I talk about here >>6455 Basically I make a base pic in a generic anime art style and then switch to the old SD1.5 based models to change the art style. This is very annoying to do because loading and unloading models takes a long time, prompting in the new XL models is very different from the old ones and all of the LoRas and VAEs from the SD1.5 models don't work in the XL ones, so to switch back and forth because I have to change everything I have loaded in addition to the model itself. Your post made me finally get off my ass and try doing exactly that, here are the results. I like that I'm able to make more complex pics than just 1girl posing, but I'll have to find some way to make the process less annoying, because just these two pics basically took a whole day to make between all of the model switching, inpainting, and even a lot of manual painting with a mouse.

Anonymous 08/03/2024 (Sat) 16:37:37 No. 6731

>>6730 Sad to see shit's got all fucked up. You should post in the /v/ AI thread since the issues you encounter effect more than just loli and could be useful info for everyone. I finally got a real computer and almost thought I could make something half-decent to satisfy my need for loli versions of characters, but I fucked something up along the way and couldn't figure out how to fix it. Here's what I ended up with, and here's my posts on the subject >>>/ais/4003 >>>/ais/4004 >>>/ais/4007 >>>/ais/4009 >>>/ais/4010 >>>/ais/4011 >>>/ais/4012 >>>/ais/4016 >>>/v/988620 I still have over 600GB of loli models, but I don't know which ones I actually need, nor do I want to trial and error my way though that massive mess when I'm not even competent to determine if I'm using a model right. I gave up, moved on to text generation, and gave up on that too when I started encountering a persistent issue with the AI dropping the second person perspective.

Anonymous 08/03/2024 (Sat) 23:44:36 No. 6734

Anonymous 08/03/2024 (Sat) 23:46:17 No. 6735

Anonymous 08/03/2024 (Sat) 23:48:15 No. 6736

Anonymous 08/03/2024 (Sat) 23:48:58 No. 6737

Anonymous 08/04/2024 (Sun) 05:13:30 No. 6739

>>6730 >people don't train models on artist names anymore. They got tired of artists claiming copyright on their art style (something they've never owned intellectual rights to). This way it is also harder to prompt the AI into recreating one of those artists actual works: trying to force the AI into copyright infringement (the IP version of posting CP to report it).

Anonymous 08/04/2024 (Sun) 15:09:29 No. 6744

Anonymous 08/04/2024 (Sun) 16:13:24 No. 6745

Here are some more pics using the art style switch method. They are certainly more laborious to make compared to the old way if doing it, but I'm decently happy with the results. >>6731 Right now I'm doing a lot of model switching, I start with loliDiffusionV1.1.0_PDXL_Lx7 for the base pic, then I use Based64mix-V3 in img2img to switch art styles, this takes a couple of tries and mashing parts of the results together in krita, then I do the upscaling also with Based64 (sometimes I vary the model depending on what the composition of the pic is). When I have a high resolution pic, I fix the defects with inpainting using mostly Based64, but sometimes I switch to XL models if I have to fix something complicated because they are smarter in recognizing things like hands or specific objects, so for inpainting I'm constantly switching between Based64, tunixStylizedPony and dreamweaverPony25DMix. I also like to do the hair in a hyper detailed style like in the 4th pic, to do that I switch to cardosAnimated_v10, load the add_detail LoRa at 2 strength and select the hair for inpainting, depending on the results I might do a second pass with a different model to blend it in better. After all of that is done I do some color balancing and contrast tweaking in krita to get those sexy reddish skin tones that I like from Boris Vallejo's art style. I think what most people miss about using AI is how powerful a tool inpainting is. It basically allows you grab a pic and flawlessly blend anything you want into it. You can fix almost anything with inpaitning, so even if you start from a shitty base pic, but you like the composition, you can gradually turn that shitty pic into something amazing. >>6739 I remember seeing some of that a while back, some guy was using img2img to slightly alter a pic he made and then posted the comparison to twitter claiming that the AI stole his pic.

Anonymous 08/04/2024 (Sun) 23:08:25 No. 6747

Anonymous 08/04/2024 (Sun) 23:09:30 No. 6748

Anonymous 08/05/2024 (Mon) 02:47:02 No. 6750

I think it's time for a new thread. Every time I try to reload the page, I get the "could not reload captcha error" and most of the files fail to load. I even sometimes get "429 slow down asshole" just from this thread. >>6745 >I think what most people miss about using AI is how powerful a tool inpainting is. It basically allows you grab a pic and flawlessly blend anything you want into it. You can fix almost anything with inpaitning That's what it seemed like when I used to easily fix the arm of the redhead in >>6731 , but when I got around to using inpainting on 2B, nothing seemed to work anymore, and it was a struggle just to do small fixes. The final pic is missing some removed black lines from the hair that I forgot to save, and those took way to long.

Anonymous 08/05/2024 (Mon) 13:46:13 No. 6752

I think I'm getting the hang of this, here are some more pics with the new method, again trying for more complex scenes. >>6750 I'm not having issues loading the thread, but yeah it's gotten pretty big, I'll have to ask the mods if I can make a new thread while this one still hasn't hit the post limit though. As for inpainting, if the AI keeps misinterpreting what's in the target area, it's usually because it's lacking the context of the bigger pic, see the tip about "context points" here >>5389 Also, what's your denoising strength when inpainting? And what's your inpaint area setting? Because that should almost always be on "only masked".

Anonymous 08/05/2024 (Mon) 20:00:31 No. 6753

>>6752 >I'll have to ask the mods if I can make a new thread while this one still hasn't hit the post limit though. If you want to make it just make it. Or hell, claim the board yourself. /loli/ is abandoned because the BO just squats the place until someone tries to claim it, at which point, he's contacted by the admins to see if he's still alive, moderates for maybe a couple weeks at most, and then fucks off again.

Anonymous 08/05/2024 (Mon) 20:09:33 No. 6754

>>6752 Damn, you weren't kidding about that scene complexity. This is some good shit.

Anonymous 08/07/2024 (Wed) 12:38:51 No. 6757

NEW THREAD >>6756 >>6756 >>6756 >>6756 >>6756 >>6756 >>6756 NEW THREAD

Quick Reply


Sage Bypass Check