[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/aco/ - Adult Cartoons


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


Infinitely generated porn right in the palm of your GPU

Last Thread: >>6999848

>>> The one link to rule them all <<<
rentry.org/sd-mashup
^ almost everything you will need is linked here, for beginners and advanced users.
Check that link before asking noob questions.

>> Quick and easy Google Colab running the latest Web UI <<
colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb
^ Command-line arguments already set for the best experience, includes a VAE, literally idiot proof, specify a path to your models if you honk, just click the play button to run each cell, and click the gradio link in the final output.

>> Washed out colors? Get a VAE <<
The "one link to rule them all" above has some in the models link here rentry.org/sdmodels
^ Place them in the VAE folder for webui, then you can select one in the settings tab.

>> Getting empty black images? <<
Add this to your command-line arguments: --no-half-vae

>>> More useful info <<<
github.com/AUTOMATIC1111/stable-diffusion-webui/wiki
>>
File: 1646794116538.png (1.1 MB, 1024x1024)
1.1 MB
1.1 MB PNG
>>
>>7008101
We making black girls?
>>
File: 1660767919903.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
>>7008123
Just wanted something different from my norm.
>>
Anyone have any diffusions of women at gloryholes?
>>
Whoever made this whats the model and prompt? Do you have an embedding of this character?
>>
>>
>>7008124
hawt
>>
File: 00031-1700645653.png (1.88 MB, 1024x1536)
1.88 MB
1.88 MB PNG
>>
>>7008123
Who's ready to get saved?
>>
>>7008311
>>
File: 1673985802020204.jpg (31 KB, 360x640)
31 KB
31 KB JPG
It's surprisingly difficult to create "low quality" selfie pics
>>
File deleted.
>>7008213
You might find that you get better results at lower resolution (512x768 is my personal favourite) with HassanBlend.
Also try adding
>high detail RAW color photograph
at the start of your prompt if it's consistently low quality / less realistic than you want.
>>
>>7008342
decided to inpaint that one
>>
File: Final 3.png (3.31 MB, 1740x1152)
3.31 MB
3.31 MB PNG
>>
File: Final final.png (1.74 MB, 1152x864)
1.74 MB
1.74 MB PNG
>>
>>
i really do find the xy grids fascinating
>>
>>
File: 00115-4090762085.png (785 KB, 640x1024)
785 KB
785 KB PNG
>>
>>
>>
we like bread dont we folks
>>
>>
>>7008926
Very pretty!
>>
First time posting in this general. Anybody know how to get the best results out of dezgo? For realistic and anime styles? I know it's not the most advanced thing, but I'm just playing around with it on that since I have a potato PC. I at least want to make up some better stuff if it's possible.
>>
>>
>>
>>
>>
>>
>>7009063
idk about dezgo but i'm having a pretty solid go of it on Colab and A1111. You can do a lot with the base SD 1.5 but you might want other models for more niche output (i.e., erotica). Someone else feel free to correct me if there's a better option but I format my prompts the following way:

1. overall composition and style (e.g. professional portrait photograph, landscape oil painting, concept art).
2. subject (nude princess, beautiful redhead woman, lineup, 1_girl). I think the consensus is that you the underscores only matter when you're using a model trained on booru tags. You can add action here too (baking bread, kneeling (on bed), walking in woods, doing pullups)
3. Details. Huge breasts, topless, blonde hair, lace choker. Different models require different details. If you prompt nude female with models trained on them, they'll know to put bobs and vagene without you prompting. This is important because each thing you prompt takes attention away from other things, which can matter for the details.
4. Style. Some people put this first. By Matt Groening, by Pablo Picasso, by Alessio Albi.

Other things: the size of your output image matters a lot. You will get very different results at 512x512 than at a portrait or landscape orientation.
I should also warn you that this is highly addictive and can fuck heavily with your brain.
>>
>>
File: image (2).png (3.45 MB, 1280x1728)
3.45 MB
3.45 MB PNG
>>
>>
>>7009155
I'll look into Colab and A1111 if I decide to go any further, but I'll try to apply your formatting to my prompts to see what I get. Thanks a bunch, friend.
>I should also warn you that this is highly addictive and can fuck heavily with your brain.
Already been fucked for a long time, dude.
>>
>>
>>
>>7009172
>Already been fucked for a long time, dude.
I mean, like, physically fucking with your sleep and having you waking up at 7 AM to proompt. seriously this is a new type of thing we're dealing with
>>
sometimes i get better results at 40 steps and sometimes i get worst results at 120 steps
it is the prompt or the model fucking up??
>>
>>7009194
Oh, well, my sleep schedule is already fucked. So it'll just be another addition to my routine.
>>
>>7009200
it's a sampler vs CFG thing, do an XY plot to find your sweet spot
>>
This shit is really useful if for example, you get a set of 3D images, lets say from a model I made in virt a mate.

Then use image2image to spice them up just enough, dont even need to make things up. With low denoising.
>>
>>7009204
im kinda retarded what do i do with xy plot i know what it does but dont know what to do!
>>
File: image (2).png (3.48 MB, 1280x1728)
3.48 MB
3.48 MB PNG
>>7009210
select xy plot from scripts, from one of the pulldown menus select CFG scale and put like 5, 7, 9, 11 - whatever your values separated by commas. from the other select sampler and type your sampler names selected by commas
>>
I really cant upscale anything, it just says I am running out of memory. I am going to research this. I only have 6gb VRAM but come on.
>>
>>
i still use hassanblend 1.3 sometimes with even 15 steps can make decent shit
>>
>>7009230
im on 2gb vram and if you make an 256x256 then upscale it to 512 or 768 it work sometimes an i get better faces
>>
>>
>>7009254
I have yet to go 32 gb of normal ram, only 16, but I have 6GB vram gtx 1660 super.

I see a lot of conflicting answers googling stuff.

I see people with 8gb VRAM also having issues.
>>
>>7009257
yes most of the time i read about xformer (never worked well on my gpu) to commands and some updates...
>>
>>7009256
alternate
>>
>>7009263
I mean I could try 256x256

In fact I going to try a different webui stable diffusion called basujindal that apparently works and uses way less vram
>>
>>
>>
For now this is just a funny little side show, I will probably forget it in about some months, even after learning about how it works.

But is a nice incentive to get a powerful pc in the net few years.
>>
>>
>>
>>7009167
You know there's a couple shibari ckpt on civitai...
>>
>>7008319
off brand liz hurley
>>
>>7009155
to add onto this anon, using an existing photo and manipulating it using SD is an effective shortcut to any photorealism issues
>GPU limit reached on colab
sad reee
>>
File: 1652207664865.png (855 KB, 640x1024)
855 KB
855 KB PNG
Messed around with img2img today.
>>
File: 1647616868900.png (239 KB, 320x512)
239 KB
239 KB PNG
>>7010348
Original for comparison
>>
File: 00173-2611464297.png (1.03 MB, 800x1024)
1.03 MB
1.03 MB PNG
>>
File: 00174-1153467261.png (1.1 MB, 800x1024)
1.1 MB
1.1 MB PNG
>>
Guys does anyone know how to assign certain color to a certain piece of clothing, hair etc? Like here i'm trying to get Daphne from Scooby-Doo in a violet bodysuit, lime green scarf and a white headband, but it return me some random assigned colors... i've tried the (()) and a selective color prefix, but nothing helped
>>
>>7010453
The same way you get it to apply specific hair colors to two different girls. Luck.
>>
>>7010457
Can you please share a guide on that? Most of the time I just generate tons of pics until i get different hair colors... i'm a newbie man, I don't know much and can't say i'm some 200IQ master mind to understand shit from vague info as well
>>
File: 1671367155428.png (1.59 MB, 1024x1536)
1.59 MB
1.59 MB PNG
>>
>>7010478
>plz write me a guide on mastering luck
>>
File: 00044-1346954708.png (1.58 MB, 1024x1024)
1.58 MB
1.58 MB PNG
>>7010502
oh shit i'm sorry, i thought your "Luck." from the first reply had meaning "good luck finding out" coz sometimes people write "Luck" instead of "Good Luck". My bad anon, i'm braindead by birth, my mom not only fell down the stairs while she was pregnant with me, she also was cheating with this huge African Doctor in my city in Germany, so his dick was so long that he would punch my head with it each time he went in...
>>
>>7010478
>>7010519
Honestly, like other anon says its a lot of actual luck involved; you're sort of at the mercy of the AI in that regard.
It's not often I try and get multiple subjects in a single image, but when I do I tend to write " multiple different girls " or something similar. Your mileage may (will) vary.
>>
>>7010590
ye, i got that, it also seems like after certain amount of tries the AI just completely ignores your (()) stuff at all and in order to make SD do your "bidding" you need to either relaunch the whole thing, or change CFG/Denoise scales by a lot, like from 7 to 1 then to 12 then back to 7 and it will start doing normal stuff again
After 32 pics got something accurate
>>
File: 2.png (1.06 MB, 640x1040)
1.06 MB
1.06 MB PNG
Call me mentally ill, but I've taken to turning people I know into well-endowed 1950s newswomen.
>>
File: 3.png (938 KB, 704x1027)
938 KB
938 KB PNG
>>7010630
>>
File: 4.png (1.35 MB, 704x1066)
1.35 MB
1.35 MB PNG
>>7010632
>>
>>7010603
Did you try a simple red hair or whatever color you wanted instead of that selective color thing?
>>
>>7010630
with what prompts?
>>
File: 6.png (1.07 MB, 704x1050)
1.07 MB
1.07 MB PNG
>>7010640
Here's the word salad I came up with in my dementia-addled brain:

1950s, 50s, old footage, black and white, (news anchorwoman:1.0) <insert whoever the fuck you want here> on the evening news, suit jacket, (cleavage, giant boobs, saggy tits, torpedo tits, huge breasts:1.1), (lips:1.2), high quality

Naturally the likenesses come from my own trained model (based off of v1.5), then I merge with g-anonmix/a couple of the more realistic porn models at various levels. So your mileage may vary depending on what model you're using
>>
File: 00208-521692806.png (982 KB, 800x1024)
982 KB
982 KB PNG
>>7010636
well the hair color is not the problem, because Karen Gillain already has red hair. I'm trying to specify the outfit colors, like purple dress, white headband, green scarf, but it always swaps colors, like instead of green scarf it will give me purple scarf and green dress etc.
>>
>>7010667
Nice! Thanks for the reply. They look good.
>>7010676
Well, did you try a simple purple dress with a lime green scarf instead of those selective color prompts? Does that make any sort of difference? If that doesn't work what about just painting the color you want over the parts and then doing img2img with a low, what is it denoise that controls the variation on the image?
>>
File: 123456.png (423 KB, 512x768)
423 KB
423 KB PNG
>>7010630
oh anon, believe me there are people more fucked than you are... like i'm turning my university groupmates into a hot sluts. You can find a video on youtube called "I faked JENNIFER LAWRENCE to educate you - Stable Diffusion Inpaint Tutorial"
Thats my asian groupmate after i was done with her. In IRL shes fucking flat as a chopping board, but her face is hot, so I've combined the two
>>
>>7010688
Yep i did, i mean thats the whole reason why i'm using "selective color" prefix now, wouldn't say it has huge difference but it gives some accuracy buff, maybe like 20-30% more accurate, coz with normal red hair, purple dress it often makes those colors dominate not on the character but on the background. About img2img stuff, well i guess thats the only way to start doing it, but it takes super long, maybe there are good img2img color models
>>
File: 1647876322658.png (932 KB, 1024x1536)
932 KB
932 KB PNG
>>
File: 16361.png (829 KB, 640x960)
829 KB
829 KB PNG
>>
File: 00246-909991671.png (966 KB, 800x1024)
966 KB
966 KB PNG
>>
File: 00262-1177886001.png (844 KB, 800x1024)
844 KB
844 KB PNG
>>
>>
ngl Kristen Stewart would make a good Harley Quinn, even in this sexy outfit she still looks like a lesbian who just came out of a prison...and she was the one doing all the fist fucking
>>
>>7010950
>attractive woman would make a good X
no doubt

>(big ass)
can recommend to try "pawg"
i've used it on huggingface/dezgo and it usually came with better results in comparison
>>
All this power and people aren't regularly making elves...
Also new version of PFG as a safetensor for you guys:
pixeldrain dot com slash u slash e34ZAKQK
>>
>>7010958
im using hassansblend 1.5 so its one of his key words to trigger the nsfw context. But ty, i will certainly try it. Often if i want pawg i just write "curvy body" and it works, its just that not writing "big ass" returns Kristen's OG butt, which is...let's be honest...looks like she was riding a horse without the saddle for 10 years straight.
>>
>>7010967
don't worry anon, this power will only grow, but the copyright for using AI to generate celebrity porn and fakes is gonna come much sooner, so people are enjoying "illegal" stuff, before they will get behind bars like Shadman and others...
>>
>>7010950
What? She would be the worst possible Harley. Harley has an insane manic personality, she's not some sort of inexpressive robot.
>>
>>7010990
I think you've seen only bad movies with her, check out "The Runaways 2010", Kristen is a GIGA level method actress. Its just people give her shit roles...look at Christian Bale as Gorr the God Butcher, method actor given shit script and shit dialog = nothing he could've saved.
>>
>>7010999
Nice
>>
File: Avatar.jpg (380 KB, 1600x1200)
380 KB
380 KB JPG
Noob question here. Can I run multiple models at the same time, or do you have to have just one for best results?
>>
File: 00171-693890332.png (901 KB, 800x1024)
901 KB
901 KB PNG
>>7011041
search for "how to blend models in Stable Diffusion" or w/e UI ur using. Also go here
https://github.com/eyriewow/merge-models
it will teach you on how to merge several models. You can also search "your model 1 and your model 2" blend maybe theres already a merged version that you can just download
>>
>>7011064
Thanks man
>>
I am a milf enjoyer
>>
Like there's just something about a curvy milf that makes the dick agree, don't you think?
>>
File: Architect.jpg (983 KB, 2304x2304)
983 KB
983 KB JPG
>>7011226
Take the architectpill
>>
>>7011265
HOW ANON
>>
>>7010970
>Hunter Biden's tattoo
wtf
>>
>>7011265
I took the mommypill. I've been refining my nipple techniques for days now

But those look good!
>>
>>7011298
With F222

(<Pierce Brosnan> :1) with a <naked fat lady>, (obese:1), BBW. drawn by <Edmund Leighton> and <Casey Baugh> and <Michael Garmash> and <Donato Giancol> and <Greg Rutkowski>, (photorealistic:1.2).
Negative prompt: Deformed, blurry, bad anatomy, disfigured, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, child, (back turned), ((mutated hands and fingers)), mustache, fupa.
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 38, Face restoration: CodeFormer, Size: 768x1152, Model hash: 44bf0551, Denoising strength: 0.3, First pass size: 512x768
>>
>>7011298
The prequel
>>
Civitai is dead?

https://civitai.com/
>>
any sites people are sharing there work on?
>>
>>7009658
>>7009665
prompts for these?
>>
>>7010630
>>7010632
>>7010633
are they German?
>>
what models do you all use?

>>7008289
>>7008319
catbox?
>>
File: 5.png (1.4 MB, 896x1000)
1.4 MB
1.4 MB PNG
>>7011682
No sir, I come from the great snowy lands of Canadia
>>
the “deliberate” model on civitai is no joke the best model i’ve ever used, just deleted 80+ GB of merges cause this surpasses all of them
>>
>>7011657
https://files.catbox.moe/jci3f0.png
https://files.catbox.moe/g3xhwx.png
>>
>>7011734
How so?
>>
>>7011265
awwyeeh. ive got my girl to ab out the first picture so far. its the best
>>
>>7011837
>ab out
what did he mean by this? act out?
It's what I keep saying, SD is an amazing source of inspiration for making "real" art.
>>
File: xavier.jpg (183 KB, 718x619)
183 KB
183 KB JPG
Would you?
>>
>>
>>7011786
it has the best handle on anatomy i've seen yet. very robust and versatile and i get extremely good results on the first try most of the time
>>
>>7010479
Very nice. Mind sharing the prompt?
>>
>>7011311
nice. model/prompt? have this milf in return
>>
>>7010712
I feel like these Karen Gillian's look more like justaminx and I find it somewhat disturbing.
>>
File: 1663688648206.png (1.29 MB, 1536x1024)
1.29 MB
1.29 MB PNG
>>7012029
Sorry, that was img2img using a sketch. This is the model https://civitai.com/models/2220/babes and for the style, (redshift style:1.1), (copeseethemald style:0.8)
>>
>>7012087
Thanks.
>>
>>7012036
can you upload yours to catbox?
>>
>>7012106
prompt is a mess and i don't even remember what the model is. i think it's 50/50 bstaber+SF_EB_SD50

https://files.catbox.moe/sxqdy8.png
>>
>>7012120
thanks. i really liked it.
>>
anyone know any words i can use to basically get some big booty spilling out from the top of a pair of jeans? i've been trying like, visible ass crack, jeans around ankles, but all it does it make things like booty shorts, skirts, or ripped jeans/normal jeans even after putting all that in the negative prompt
>>
>>7012137
Skindentation will put in a lot of work, and maybe Bursting_Ass depending on the model.
>>
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
RuntimeError: CUDA out of memory. Tried to allocate 512.00 MiB (GPU 0; 8.00 GiB total capacity; 5.64 GiB already allocated; 226.75 MiB free; 6.27 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Can someone please explain to me why I get the above error when I try to train a hypernetwork?

It didn't do this pre update and I've tried every way I know to fix this, I even went so far as to reformat my whole computer and reinstall SD.
>>
>>
>>7012350
Have you tried 'git checkout' to a previous to rollback?
>>
can someone help me to resolve the issue im getting? Yday everything worked alright, now i have this and pressing any key just closes the window...
>>
>>7012774
Only thing I can suggest is removing the arguments (med-vram etc) and trying again
reinstalling if that doesnt work maybe
its kinda hard to tech support issues with the webui, IMO
>>
>>7012774
Maybe the model file was not downloaded completely or is corrupt. Try redownload, or remove it and rename a model in there as "model.ckpt" so it will load it as a default.
>>
File: 00003-899773050.png (1000 KB, 1024x768)
1000 KB
1000 KB PNG
>>
>>7012814
>>7012816
in case you incounter same stuff, just go to your folder with SD and start git and write git pull, it added some files and deleted others, prob model was already updated but not the web ui
>>
File: snip.png (3 KB, 240x195)
3 KB
3 KB PNG
>>7012822
You know, I was actually going to suggest doing that after I posted >>7012814

I just put
git pull
in my webui-user.bat so it auto updates when i open it
>>
>>7011265
Bloody brilliant.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>7009204
>>7008941

wwhat model guys?
>>
>>
>>
>>7008892
>>7008765
catbox?
>>
>>7013060
mix of the original dreamlike and f222 at 50-50. this one is either the new dreamlike or protogen i forget which
>>
>>7012036
>Model
Custom merge of UPM and HasDX

>Prompt

>Positive: 1girl, In the style of Norman Rockwell, Victorian art, painting, By Artgerm and wlop, Michael Garmash, Justin Gerard, A mother in the 1950s wearing a nightgown, digital painting, 8k, HDR, highres, 1girl, mature_female, perfect face, perfect gorgeous eyes, curvy, large nipples:1.5, gigantic breasts, thick thighs and wide hips:1.2, areolae, puffy nipples:1.2, full body, SFW, diaphanous, see-through, nightgown, nightshirt, lingere, detailed clothing, GILFL:1.2, portrait, detailed skin, old face, slight wrinkles1.1, tanlines:1.2, pale skin:1.2, pale, PAWG

>Negative prompt: 2girls, 2women, 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

>Steps: 22, Sampler: Heun, CFG scale: 8.5, Seed: 1918143751, Size: 768x768, Model hash: a8c08573f6, Batch size: 2, Batch pos: 1, Denoising strength: 0.66, Mask blur: 4
>>
Inshallah brozzers I am horny for BIG HIJABI MOMMIES

>>7009155
>>7009224
I love your proportions on these >>7009237 >>7009256 >>7009269
>>
By Allah this is haram!
>>
>>7013299
hey, anon. i've just started messing with this recently. i couldn't find the UPM model you said..
>>
>>7013318
here you go brozzer
https://civitai.com/models/2661/uber-realistic-porn-merge-urpm
>>
File: lara_earns_it_c2 copy.jpg (521 KB, 1176x1536)
521 KB
521 KB JPG
Fixed the color on this and added more cum because why not?
Were colors desaturated washed out because no vae? I put one on, but can't remember when or if it goes with the model I used.
>>
>>7013299
nice
>>
>>
>>
>>
>>
>>
>>7013321
>>7013316
>>7013309
>>7013299
>>
>>7013462
>>
>>
>>7013462
full prompt?
>>
File: 1659831038495.png (1.59 MB, 1024x1536)
1.59 MB
1.59 MB PNG
>>
>>7013438
>>7013480
>>7013440
The face is too distorted. You can get better results with this:
https://civitai.com/models/4631/taylor-swift
>>
>>7013488

its an older hassan blend dude i know but thanks for the link! i will try it
>>
>>7013321
>>7013299
the best one yet
>>
File: 1643059597069.png (1.54 MB, 1024x1536)
1.54 MB
1.54 MB PNG
>>
Models seem to really have a lot of trouble with nipples and even puffy vaginas.

No doubt in a few years it will be perfect.
>>
>>7013548
i don't know about the anons here, but it's not like all the images the a.i do are perfect. maybe i'm incompetent. i only get it right 3 out of 10 attempts.
>>
>>7012120
whats SF_EB_SD50 and where can I get it?
>>
>>7013553
i have a pretty high hit rate when i'm going for simple portraits but when i try to do more complex compositions it goes down significantly
>>
>>7011783
>model 8f4a0a1e
what mix is this?
>>
>>7013553
Oh yeah sure.
I am just saying we are seeing the tip of the iceberg.
>>
>>7013675
50 % AbyssOrangeMix2_hard and 50 % bstaber
>>
File: 00002.png (2.48 MB, 1536x1536)
2.48 MB
2.48 MB PNG
>model has multiple tokens for pissing
>insert them all with emphasis
>still can never figure it out
>>
>>7013627
smirkingface on patreon (its free)
>>
>>7013750
What is Bstaber? Is it a model, lora, etc?
>>
>>7013914
It's a model

magnet:?xt=urn:btih:5c151e2b6f48d3b1bb6d8f390487d8b7601cf81b&dn=Bstaber.safetensors
>>
>>7013917
What kind of model anon? I just downloaded it from here >>7004129 >>7004129
Are those the same things?
>>
>>7013921
Yep, that's the same mix. You don't have to download the torrent I posted the magnet link for, just get the file you linked and you're set.
>>
>>7013934
What is in it? It seems almost kind of shady anon
>>
>>7013938
I don't know. Why would it be shady?
>>
FBI is tracking Bstaber users
>>
Any tips for improving the eyes/faces? Any general 2D tips are welcome
>>
>>7013968
Not unhappy with the results so far however...
>>
>>7013953
FBI competent and proactive, news at 11
>>
File: 00277-3200350248.png (836 KB, 640x960)
836 KB
836 KB PNG
>>7013953
what about bstaberx?
>>
the general problem with this tool is that the lack of compositional control means that it's hard to tell a story in your pictures. i like this one because it's like her friend is keeping her company while she's being publicly humiliated
>>
>>7014026
Its certainly is something that will improve. If nothing else image2image is already plenty powerful.
>>
>>7014032
hopefully there will be a model that understands natural language prompts a bit better
>>
>>
Getting this strange pink tone on some of the skin
>>
File: 00006.png (751 KB, 1024x1024)
751 KB
751 KB PNG
i think i'm getting better
>>
>>
File: 00008.png (872 KB, 1024x1024)
872 KB
872 KB PNG
>>7014191
can some anon instruct me on what upscale I should use in real life photo prompts? i think i'm loosing details with the SwinIR 4x..
>>
>>7014191
>>7014209
where's her head lol i use r-esrgan 4x at .5 denoising because latent adds details like random hands everywhere
>>
>>7014256
>where's her head
how can i make her always displaying full body? "full body"?. maybe sometimes it doesn't work.
>r-esrgan 4x at .5 denoising because latent adds details like random hands everywhere
i'll try it. thanks.
>>
File: IMG_8316.jpg (1.09 MB, 3124x2766)
1.09 MB
1.09 MB JPG
Made some stickers with novelAI generated art
>>
>>7014085
There's some bug on automatic111 with the VAE that gives you pink/purple bruising. Reload the VAE and apply it. Not sure if that's your problem, but try it with and without the VAE.
>>
>>7014278
change your dimensions. i have good luck at 576x768. you can also take the headless image, send to img2img, and use the outpainting mk2 script with only "up" checked and the exact same prompt.
>>
Don't usually post my celeb's but this one came out pretty damned good.
>>
>>7013990
>>7013968
just inpaint the face, use the original resolution settings, set denoise to 0.56, cfg 7, original, inpaint masked padding 84
>>
File: 1664575431624-1814436937.png (2.37 MB, 2048x768)
2.37 MB
2.37 MB PNG
>>
i was expecting a bdsm mask, but it did well
>>
>>
>>
>>
File: 1646680849164.png (1.47 MB, 1024x1536)
1.47 MB
1.47 MB PNG
>>7014209
Download the 4x-Ultrasharp from https://upscale.wiki/wiki/Model_Database and place the .pth inside the ESRGAN folder stable-diffusion-webui/models/ESRGAN

Download https://github.com/Coyote-A/ultimate-upscale-for-automatic1111, you can use the extension tab. For the settings, use

Steps - 40
CFG - 4
Denoising strength - 0.25

Script - Ultimate SD Upscale
Target size type - Scale from Image Size 2
Upscaler - 4x-Ultrasharp
Type - Chess
Tile size - 512
Mask blue - 16
Padding - 32
Seam fix - None
Save Options - Upscaled

If you want more details, keep increasing the denoising strength by 0.05 until satisfied. This should work with most images.
>>
>>
>>
>>
>>7014594
>>7014601
>>7014603
cool
>>
>>
File: 00039.jpg (545 KB, 1440x2304)
545 KB
545 KB JPG
>>
File: Archer Nap 2.png (1.27 MB, 1080x1080)
1.27 MB
1.27 MB PNG
>spend all day making hundreds of images
>only kept 25 after coming up with creative solutions to hide shitty hands
I swear 98% of what I make is utterly unusable because of hands.
>>
>>
File: grid-1111.png (1.6 MB, 1024x1024)
1.6 MB
1.6 MB PNG
>>7014332
didn't even know about VAE or even have one but downloaded and playing around, cant get the pink to go away still, thank you for the info. I'll keep trying
>>
>>7014627
looks great
>>
File: Untitled-1.png (3.02 MB, 1280x1920)
3.02 MB
3.02 MB PNG
Comparisons between the three version of my Decker Loras used on my Miracle Max's Mix v10.
>>
File: 1648957980920.png (1.49 MB, 1024x1536)
1.49 MB
1.49 MB PNG
>>
>>7014643
all those were the same seed and the simple ass prompt:
nsfw, Rias Gremory
added a negative prompt:
"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"
and the pic went downhill. thats a first
>>
>>
>>7014594
>>7014601
>>7014603
>>7014607
>>7014616
love these, can you catbox please?
>>
File: 00010.png (2.63 MB, 1536x1536)
2.63 MB
2.63 MB PNG
>>
File: 1659702643200.png (630 KB, 512x960)
630 KB
630 KB PNG
>>7014680
decker anon, dont know if u remember me, I had asked you for your LORA quite sometime back.. I made a decker style embed..its not quite there yet, more focussed on realism..picrel
>>
Is there anything like civitai for embeddings/loras?
>>
>>7011716
>>7010667
>>7010630
love this let's have more of this character model anon.
>>
>>7014279
Gross.
>>
>>7015314
I agree

>>7011716
>>7010667
>>7010630
Please enrich our day with more of your sturdy well endowed cathode ray tube news sirens.
>>
>>7015297
civitai has both of these, check your filters maybe
>>
>>7015332
Right, I didn't check the filters.
>>
>tfw can't do microkini real life girls
>tfw no face
it hurts so much
>>
File: Gwen 489.png (1.53 MB, 2573x2460)
1.53 MB
1.53 MB PNG
can the models do adult gwen?
maybe by making a lora, whatever that is, using image sample?

while i like her ponytail version, i dunno if i have enough images depicting her with it.

i can provide some images of her adult version
a bunch
>>
Good progress
>>
First time I proompt in 3 months.
>>
File: tile.jpg (1.27 MB, 3072x2304)
1.27 MB
1.27 MB JPG
>>7015453
>>
>>7015453
Kstew on that MCU buff bitch program
>>
Any models with Liv Tyler, or Jessica Rabbit?
>>
>>7015481 sd1.5
>>
>>7013299
>>7013309
>>7013316
>>7013321
moar
>>
File: 1671719550610086.jpg (106 KB, 765x507)
106 KB
106 KB JPG
>>7015550
Danke!
>>
>>7008319
You can crappify photos pretty easily in thinga like photoshop and GIMP
>>
>>
File: 1650902279329.jpg (472 KB, 2048x3072)
472 KB
472 KB JPG
>>7015436
>maybe by making a lora, whatever that is, using image sample?
Yes, training is the easiest way. I tried training Talia, had no idea what I was doing, and it worked. I can see signs of overfitting or whatever the name is, but it fucking worked.
>>
File: mk_1.png (1.33 MB, 1315x800)
1.33 MB
1.33 MB PNG
Protogen v2.2 (Anime) + https://civitai.com/models/4318/mila-kunis-embedding embed.
original prompt before upscaling, inpaint and manual crop:

parameters

_modelshoot style, (extremely detailed CG unity 8k wallpaper), full shot body photo of the most beautiful artwork in the world, professional majestic impressionism oil painting by Waterhouse, John Constable, Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, trending on ArtStation, trending on CGSociety, Intricate, High Detail, dramatic, trending on artstation, trending on CGsociety, (milkun) lying on a bed, 40 yo, (milf), plain magenta t shirt, magenta beanie, blue denim daisy dukes, (mature woman), round glasses, dimples, wrinkles, short hair, legs
Negative prompt: canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), wierd colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render, loli, kawaii, young, child, (jeans), (pants), t shirt logo, t shirt text, large breasts, bobble hat
Steps: 26, Sampler: DDIM, CFG scale: 7, Seed: 730577164, Size: 720x400, Model hash: 1254103966, Model: protogenV22Anime_22, Batch size: 4, Batch pos: 3, ENSD: 31337
>>
>>
>>
How often does an AI give you images that are just completely cursed?
>>
I have come to the obvious conclusion that any detail I write will be highlighted. If i highlight the breasts, it is not possible to get images with poses where the breasts are not shown. The downside of this is that it is not possible to get a diversified pose every time i run stable diffusion. Am i wrong?
>>
>>
>>7016276
I could be totally wrong, but the way I am interpreting things, is that the AI has no actual idea what breasts are or what you're asking for. It's just referencing source knowledge that contains the key words. So yeah, the fact that you mentioned breasts positively at all in any context, basically tells it to put breasts in the output.
>>
>>7016313
The AI does not directly reference any knowledge, that's artist-tier knowledge that is akin to thinking it copies and pastes existing art. Final conclusion is mostly correct, though.
>>
>>7016313
>>7016317
Yeah, you are right.. I'm describing a image after all. I'll try to use the word "fat" and filter out the other characteristics of a fat woman to get the "big breasts". I am concerned about being so strict because it sometimes makes the picture look weird or ignores several words.
>>
>>7016326
Have you tried "hourglass figure"?
>>
>>7016333
>hourglass figure
Looks like a great keyword. I will try it when I am near the computer. Thanks, friend.
>>
can i use the upscaler with 2gb vram???? always get crash or errors
>>
>>
>>
File: 1668543881987.png (1.19 MB, 1024x1536)
1.19 MB
1.19 MB PNG
>>
>>
>>7016106
Almost never unless there's just some string of words that it hates. Recently for me it was "On all fours" that made it spaz out and give me freakshows.
>>
File: 1647588088297.png (1.37 MB, 1024x1536)
1.37 MB
1.37 MB PNG
>>
File: 1643690691443.png (1.06 MB, 1024x1536)
1.06 MB
1.06 MB PNG
>>
>>7015279
NICE!
>>
>>7016937
>let me ASS you a few questions
>>
>>7016957
kek
>>
>>7013464
>https://civitai.com/models/2661/uber-realistic-porn-merge-urpm

Prompt or Catbox?
>>
Why do some models include VAEs? What are they? Do I ever need to use one other than the default one SD comes with and why?
>>
>>7014603
>>7014601
>>7014594
gloryportal
>>
>>7008097
I love this.
>>
>Want to blend human with animal parts to get mythical hybrid.
>Just end up with a human and a deformed pet nearby.
>...
>Add small reference to gremlins.
>Now everything blends together.
>Almost any two things can blend.
>Sometimes comes out cross-eyed.
>>
File: 00032.png (167 KB, 512x768)
167 KB
167 KB PNG
>>
File: 00030.png (148 KB, 512x768)
148 KB
148 KB PNG
>>
File: 00026.png (107 KB, 512x768)
107 KB
107 KB PNG
>>
>>7014614
inpaint!
no need to throw away a perfectly good image because of some bad parts; try inpaint over them!
>>
>>
Anyone doing request? My computer is fucked right now. I wanted to see Doomsday Groping power girl.
>>
Anyone figured out a good prompt to get them on all fours? Crawling and the like
>>
>>
>>
>>7019019
catbox?
>>
>>7019023
i basically stole it from /h/, but
https://files.catbox.moe/9m2sjz.png
>>
>>7019026
ty
>>
whats the difference between this thread and the one in /h/
>>
>>7019044
/aco/ is for western style cartoon images
>>
>>7019048
yes but for most of the images in these threads, could you tell which board it's supposed to be from?
>>
>>7019049
you asked i answered
im not a janny so its not up to me whats in the thread
>>
Huh. Was having some trouble with my prompt producing blurry/shaky images and went from mentioning 1 manga artist to adding 3 more that I looked okay and were highly rated on r34, and suddenly it spits out this perfectly clear and coherent image on a low sample single test run.
It looks nothing like Chun-Li but this is the best image I've ever gotten out of it.
>>
man it takes forever, but once you can get it to start inpainting on an exisiting face, then keep tweaking the denoising up and down for novelty, its addicting how hot it is to take an existing face and create novel porn on it.

I'm in heaven.
>>
>>7019075
the face looks off, it's at the wrong angle with the top of the head
>>
>>
File: Untitled-1.jpg (163 KB, 868x1010)
163 KB
163 KB JPG
>>7015314
>>7015327
How could I possibly say no?
>>
File: Untitled-2.jpg (278 KB, 927x1217)
278 KB
278 KB JPG
>>7019196
Same girl!
>>
>>7018707
That pixelated style reminded me Larry Laffer games.
>>
>>7018707
>>7018708
>>7018710

No excuse for wegs now?

Is there a bimbo model?
>>
File: Untitled-3.jpg (353 KB, 960x1362)
353 KB
353 KB JPG
>>7019201
>>
File: Untitled-4.png (1.42 MB, 960x1416)
1.42 MB
1.42 MB PNG
>>7019237
With a little bit of stylization
>>
>>7019049
its not really the style anymore but rather the content
does the character comes from western cartoon?
is the deepfaked celebrity a non-japanese celebrity?
then again /h/ doesn't tough any real person
>>
File: volpina 14.png (2.48 MB, 2560x1440)
2.48 MB
2.48 MB PNG
>>7015874
she looks like lila rossi from miraculous ladybug
probably just a coincidence

anyway, does this mean that if i share a zip file next thread, someone can generously make a lora model?
>>
File: 1657727839733.png (1.92 MB, 1024x1536)
1.92 MB
1.92 MB PNG
>>7019369
If the images are cropped to the right size and captioned, probably.
>>
had some fun creating posters, some were alright, most looked like shit though
>>
>>7019638
>JOEY ERECTION
>>
>>
>>7019787
>>
what are some negatives in order not to get anime faces? I'm using an aom2 mix
>>
Well eventually my crazy prompt did something. Alright.
Using urpm of all things.


I find more fun altering existing images than using text.
>>
File: No wm compressed.jpg (2.73 MB, 4096x2560)
2.73 MB
2.73 MB JPG
>>
>>
>>
>>7019986
Damn just noticed the hand
>>
Should I use the underline as the spaces in keywords? I dont get it. I'm getting good outputs with or without them.
>>
>>
File: Candlelit Gaze.webm (2.91 MB, 608x1080)
2.91 MB
2.91 MB WEBM
>>7008097
>>
File: final.png (556 KB, 768x512)
556 KB
556 KB PNG
>>
>>7020495
woops, thought I was on /h/ sdg. my mistake.
>>
File: 1663217030992.png (320 KB, 512x768)
320 KB
320 KB PNG
>>
>>7019969
i like that one
>>
New Thread
>>7020579
>>7020579
>>7020579
>>
>>7019787
>>7019788
which came first and which is in-painted?



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.