File: NSFW_K-0000480.jpg (55 KB, 512x768)

55 KB JPG

>>8328842
Dude, post some pics or the thread might get culled.

>>8328883
sorry got distracted
am not regular baker, will have dump ready next time

>>8328896
I understand, it happens, it was just some adive because that has happened before in the blink of an eye.
>>8328903
Really digging this one.

>>8329002
better kayas lined up in the pipeline already, need to finetune my lora settings

File: 00146-3913575962.png (793 KB, 1608x1248)

793 KB PNG

after some testing, zonkey is the go-to pony realistic model for good shadow, hard to tune, but good results

>>8328893
That's really good. Model/prompt?

File: 116775249263546856-SD.png (616 KB, 536x960)

616 KB PNG

Steinfeld is goated

>>8329123
agreed, due to all the “details” zonkey adds I had to turn the add-details-xl lora way down. switched back to pony realism since I’m going for a “covered in freckles” look that zonkey couldnt quite seem to pull off

>>8329236
is add-details any good for better skin? I ran some tests with detailerlora and that was really disappointing

>>8329246
for skin I really like the Freckles and Skin XL lora, I think it goes light on freckles if you just prompt the “natural skin” keyword. iiuc, add-detail-xl adds some noise to the initial steps that diffuse into more details as it progresses, which can help with skin detail depending on the model but isnt directly tied to more realistic skin.

File: 00488.png (2.41 MB, 1616x2000)

2.41 MB PNG

>This gym is for Congress officials only, what are you doing here?

File: 00220-3400901824.jpg (227 KB, 768x1344)

227 KB JPG

>>8329466
Are you the guy using a website that doesn't support loras?

>>8329498
No, I'm pretty new to this whole thing

File: 00231-3736463737.jpg (204 KB, 768x1344)

204 KB JPG

>>8329524
not bad gens for a newbie
you gotta use a LoRA of her though

>>8329123
What are you doing for upscaling? everything I tried to do with zonkey turned it into nasty blobs even though it looked pretty good at initial gen size.

File: 00494.jpg (425 KB, 2496x2520)

425 KB JPG

>>8329535
I'm using one but still figuring out everything. Pretty inconsistent but I don't mind her looking like a generic latina either

File: 1718902792821320img2img.jpg (232 KB, 768x1280)

232 KB JPG

>>8329561
Which LoRA is it? It looks like an embedding actually (really small files, usually bad likeness). If you change the weight, does it have any effect?
Maybe you picked the wrong one and it's not compatible with your model and silently ignored. What we are seeing then is what the base model thinks she looks like
>>8329478
This should look more like the pic
Your gens look like what it makes during the first few steps of the gen. Really smooth, young looking face which is missing the important features

File: 00496.png (3.29 MB, 2000x2016)

3.29 MB PNG

>>8329596
I downloaded some random one on civitai but I think it got removed, and I definitely see what you mean with the smoothness, it takes some effort to get more of her features but again I don't mind

File: 00481.png (1.69 MB, 1200x1352)

1.69 MB PNG

>>8329596
This is one I did of Tara Strong for a /co/ thread, seemed easier

>>8329640
yeah she looks basically like a super cute young version of her, maybe a younger sister
barely recognizable but still good looking loras like that are good for making consistent rando characters
also I'm making a note of the rock climbing theme to use for later, very creative
>>8329663
maybe missing something in the negative prompt to get a better face (the eye asymmetry is a giveaway)

File: 1716652489095104.png (2.58 MB, 1440x1632)

2.58 MB PNG

>>8328512
which model/lora is this anon?

File: 00039-3532168931.png (1.31 MB, 1024x1024)

1.31 MB PNG

>>8329778
realismenginexl, it's an img2img of a 3d render

>>8329824
are you a wizard in blender or is there a tool to make these renders for you?

File: 00254-2038614133.jpg (243 KB, 1728x1224)

243 KB JPG

>>8329461
summertime saga?

>>8329859
>>8329864
>black dots on those white shirts
you prompted for cows didn't ya?

File: 685678u567u.png (1.96 MB, 1024x1024)

1.96 MB PNG

>>8329890
I make them myself, the base image itself isn't too complicated, just enough to set up the pose and lighting

>>8329824
thanks!

File: 00147-2879245501.png (1.31 MB, 1024x1024)

1.31 MB PNG

>>8330001
it was freckles

still trying to find the fine line between expressions, randomness in facial structure, freckles, and general fuckability

File: 00142-3847751542.jpg (1.56 MB, 2080x3040)

1.56 MB JPG

>>8330435
and remembering to add the picture

File: 00192-3847751592.jpg (1.54 MB, 2080x3040)

1.54 MB JPG

that's not quite a banana but good try

File: 00026-2335929086.jpg (1.55 MB, 2400x3200)

1.55 MB JPG

alternating steps with [facial_expression1|facial_expression2] is causing wierd artifacts sometimes

File: 00044-2335929105.jpg (1.76 MB, 2400x3200)

1.76 MB JPG

and keep trying on the upside-down high heels

File: 00069-2335929131.jpg (1.92 MB, 2400x3200)

1.92 MB JPG

"collar" will often create a leash even if unprompted, might have to stick to "black leather choker" or something similar

File: 00037-68074237.jpg (247 KB, 768x1344)

247 KB JPG

>>8330442
yeah I haven't found good use for that syntax yet
alternating between clothes and nudity can sometimes give you extra skimpification but it's real fragile
for more interesting faces, I like to use different facial expressions in the base gen versus the adetailer face prompt
if the denoising strength is in a sweet spot, the adetailer run has to incorporate the first expression into the result somehow
if expressions are a bit randomized (dynamic prompts/wildcards), you never ever end up with exactly the same face twice

File: 00079-2335929141.jpg (1.9 MB, 2400x3200)

1.9 MB JPG

>>8330462
sure... I get great use out of it for certain tints and highlighted hair like
[(ginger hair:{0.4|0.6|0.8})|(red hair:{1|1.2|1.4})|(orange hair:{1.2|1.4|1.6})]
in 1.5 / XL it was the best way I found to blend celeb faces, I felt like almost every single face was attractive. The right part of the OP pic is about as close as I've gotten to what I'm looking for, but even when using that exact same face prompt for different locations/poses it doesn't look like the same type of face

File: 00067-2335929129.jpg (1.75 MB, 2400x3200)

1.75 MB JPG

>>8330486
and those goddamn big freckle clumps
< here's an example of a gen where SD just decides "fuck everything you told me about sharp focus and high resolution and detailed freckled skin, using the exact same loras and prompting as other gens that have detail - here's this shit"

File: 00009-3028033879.png (1.15 MB, 832x1216)

1.15 MB PNG

>>8329559
seconding this question; i'm having a lot of trouble upscaling Zonkey, partially due to OOM errors and partially due to me not knowing what i'm doing

File: 00185-490844986.jpg (1.53 MB, 2048x3072)

1.53 MB JPG

just went on a nostalgia browse of old 1.5 folders... now I need to purge like 99% of it, everything looks so plastic now

File: 00072-3690595455.png (1.83 MB, 1248x1824)

1.83 MB PNG

>>8330524
i deleted my entire 1.5 coom directory in a fit of pique and felt a great karmic debt lifted. unironically this might be having some effect in the spirit realm

File: 00200-2228419595.jpg (1.91 MB, 2560x3200)

1.91 MB JPG

>>8330529
the suddenly unallocated sectors on your drive suddenly freed for higher purposes contributed to the delay of the heat death of the universe
< base XL had almost no hardcore without using pose loras or control net, I always got either legs and cock with no head or torso, or disembodied floating penises

File: 00172-3477019947.jpg (1.76 MB, 2240x2880)

1.76 MB JPG

but I loved the XL faces

File: 00068-4294755404.jpg (1.93 MB, 2560x3200)

1.93 MB JPG

>>8330561
>>8330565
and it was so rare to get redhead pubes, or even pubes at all, even with heavy weighting, and as we all know redheads are better with pubes

File: 00499.jpg (921 KB, 3200x4000)

921 KB JPG

I need a new model idea or people are going to think I'm obsessed with this bitch

Whoever did Emma Watson in the previous thread, can you do a 34 year old Emma with small saggy tits, and extremely hairy forearms and happy trail please?

>>8330804
Oh yeah and if it's SFW even better so I can post her on /tv/, thank you.

File: 00000-1281645103.jpg (346 KB, 1536x2560)

346 KB JPG

Hpw do i avoid samefce on Ponyrealism?

File: zonkey settings.png (55 KB, 928x587)

55 KB PNG

>>8329559
>>8330520

Something like this

for the Vram issue, im using forge, with is a custom version of A1111 wich have a fantastic Vram management

>>8330849
gorgeous!

Can u share prompt?

>>8330970
appreciate you.

File: 00001-3602580884.png (3.97 MB, 1536x2560)

3.97 MB PNG

>>8330991
Sure
https://files.catbox.moe/t9x1js.png

File: Screenshot 2024-06-21 000148.png (1.08 MB, 1655x1041)

1.08 MB PNG

I have literally never posted here, but I'm here to tell you faggots that I'm training something that might interest you

>>8331137
Emilia Clarke doing weird sexual stuff to ScarJo?
:)))

>>8330970
After you make an image with this, what's the peak VRAM usage? (values under the image A: R: Sys:)

Outside Louis Rossman has anyone else giving general permission for AI generation using their image? I am testing PuLID and am tired of just looping something I created.

>>8331311
Unlikely.
You can always use someone whose image is in the public domain, ie someone that has been dead for over 70 yers.

Whoever is that retard posting a bunch of not AOC stuff, just stop

It's so far from looking like her it is ass

>>8331375
it's so far off that it's basically a rando character
curb your autism

>>8331311
aella lol

File: img.png (5 KB, 273x69)

5 KB PNG

>>8331375

the magic of Forge

>>8331420

>>8331374

Sorry wrong reply

>>8331269

Sorry im retard i dont know how to use 4chan see image above

>>8331420
I'm gonna have to give forge a try. I'm running out with 12GB all the time. That's even with low-vram mode (which shouldn't be needed with such a card)

>>8331453
if you're using a1111 the switch to forge is game-changing and extremely simple

File: 00066-2335929128.jpg (1.77 MB, 2400x3200)

1.77 MB JPG

>>8330849
I've been fighting it for a while. right now I've got a few XL celeb loras at mid weight, either just one or alternating between two each step, with wildcard facial expressions/"beauty" adjectives/weightings. not 100% perfect but it helps. another anon mentioned using a higher weight celeb lora in the adetailer prompt, but since wildcards don't work in that prompt and I try to steer away from hardcoding I haven't tested it much
< this one for example, I wouldnt normally upscale because of the fingers and pussy, but this was like a 0.8 emilia clarke gen
>>8330438
I remember this one was "young britney spears"

>>8331469
And it's even easier if you use stability matrix that joins all your loras, models, in a "global installation".

100% recommended

>>8331482
wildcards and all other dynamic prompt features do work in the adetailer prompt
what doesn't work is the fact they won't match up in the random parts (unless using cyclic) and you can't use variables set in the main prompt in an adetailer prompt either (there's a bug report that went nowhere)
for the alternation syntax, you can also specify more than two for even more tweaking
also do something like [one|one|two] so it sees the second one only every third step

>>8330849
>>8331482
I just had an idea. Maybe this is a good use case for ancestral samplers. You can specify a different sampler for an adetailer tab no problem
The ancestral ones (those ending in " a") are non-converging. That is, if you regen the same seed with one extra step for instance it will still make modifications instead of looking "better". It doesn't beeline for a single target image, instead it "circles" around some target.
Bad to use for an image in general, but might help break the monotonic faces. Could also help for freckles. The idea is that the normal samplers are converging to something that is washed out/blurry and these sampler might not.

>>8331482
The LoRAs are always active, you are just switching the keywords
What happens when you just invoke all the keywords at once? You can finetune their weights individually (unless they use the same, sks woman etc.)
You won't get multiple persons this way. The CLIP interrogator regularly spits out multiple persons names to describe a single one too
Remember the AI doesn't know what a person is. A name or LoRA keyword is just another attribute that you can all use together like "tall,thin,attractive" etc.
I regularly mix in the representation of a celeb in the checkpoint together with the LoRA keyword to increase likeness.

>>8331374
dead person for the win. Thanks.

>>8331394
That is a bag of crazy I didn't need to open.

>>8331453
https://github.com/lllyasviel/stable-diffusion-webui-forge?tab=readme-ov-file#forge-backend

Just a heads up about those flags.

>>8331392
Curb your retardation

File: 00509.jpg (747 KB, 4000x4000)

747 KB JPG

I just like latinas

File: 00022-4152889425.jpg (1.9 MB, 2400x3200)

1.9 MB JPG

thanks for the face advice guys, need to do some more tweaking - in the meantime I have a temporary fix

>>8331375
beggars can't be choosers
if you actually want AOC, that's the worst way possible to make a request here

>>8331869
>>8331871
>>8332087
cool stuff

>>8331997
I see a pattern here today

>>8332113
perfect. did you prompt for transparent clothes? never seen them with wetness before, wow

File: 21880-53113283.png (2.45 MB, 1344x1728)

2.45 MB PNG

"Anon... We need to talk about your grades lately."

>>8332355
realistic loli are CP you sick fuck

File: 00009-1112136877.png (3.69 MB, 1440x1824)

3.69 MB PNG

>>8331375
this is why celeb posting should be discouraged

>>8330970
thank you!

File: 00026-3013063060.png (1.16 MB, 832x1216)

1.16 MB PNG

>>8330970
yea i'm in forge but i oom with these settings, this in unique to Zonkey and i can upscale much bigger with other XL models

>>8332749
instead of giving out posting advice, try genning some bondage that isn't shit for a change
I remember your stuff from a few threads back with the three girls in crappy restraints

>>8332764
zonkey is always more VRAM hungry for some reason
it looks really good with another realistic pony model as a refiner
I wonder if that step can be skipped and using img2img latent upscale in its place
then zonkey VRAM on upscale wouldn't be an issue since the other model will be used for that
not sure if that compromises quality. also a manual process

>>8332773
ok make a request then

File: 00084-2335929146.jpg (1.74 MB, 2400x3200)

1.74 MB JPG

>>8332181
pony [realism] uses danbooru tags, so things like "see-through [clothing]" and "transparent [clothing]" work well, "wet" tends to cover the skin in running water if you aren't careful with the weight. I also use "covered [body part]" for fabric over nipples/pussy etc with the outlines poking through, like >>8332113

File: 00515.jpg (517 KB, 3200x3200)

517 KB JPG

Something about AI armpits is just perfect, and I've never really been into them before

File: 00100-1493505850.png (1.87 MB, 1160x1496)

1.87 MB PNG

>>8333675

>>8333675
>>8333707
More like Pooters

>>8333135
Looks like you're using "face restore". Don't. It'll compromise the likeness considerably.

File: 00116-1609412315.png (1.91 MB, 1496x1160)

1.91 MB PNG

>>8333720

>>8333737
Loving it, any chance you can add some redness or a hand print?

File: 00149-1609412315.png (2.22 MB, 1608x1248)

2.22 MB PNG

>>8333751

like this? (first time doing this)

>>8333787
Good start

File: Comparision.png (3.7 MB, 6242x1608)

3.7 MB PNG

New testing with zonkey, what do u think guys?

>>8331997
how much vram do you have, you focking bastard

File: 00167-3893001560.png (1.37 MB, 960x1280)

1.37 MB PNG

>>8328842
where do i post similar pictures, here or in /d/, i don't understand.
/d/ has more futa, but cartoonish, here is more photoreal, but almost no futa

>>8333966
is long as it's realistic, I'd say go for it
there's lots of stuff that would belong here but isn't in the thread, there's just not enough people genning
just no realistic loli, cause that's outright CP

File: 00081-770197927.png (1.48 MB, 960x1280)

1.48 MB PNG

>>8333976
ok

File: NSFW_K-0000250.jpg (47 KB, 512x768)

47 KB JPG

>>8333966
>>>/d/ or some dedicated futa thread.

>>8334006
dedicated celeb thread first

>>8334028
>>8334035
>>8334053
>>8334065
>when you order Daisy Ridley on Wish

File: 22069-2637742698.png (2.6 MB, 1344x1728)

2.6 MB PNG

>>8333941
My vote goes to hiresfix+adetailer.
Personally the only adetailer i use on zonkey is mediapipe_face_mesh_eyes_only.

File: 22070-4116126851.png (2.39 MB, 1344x1728)

2.39 MB PNG

>>8333966
/aco/ is /d/-lite.
Just look at the catalog.

File: 00518.jpg (647 KB, 3188x4000)

647 KB JPG

>>8333734
You are correct, I guess I still have a lot to mess with in the settings.

Also, I know this ain't a democracy but I vote NO to futa stuff. Otherwise I'm going to have to generate a lot more non-futa to post

>>8334461
same. once you let in the futa, other futa users start posting, then the topic is basically only futa, then the furries come in, then the original topic is non existent

In a sense, the futa stuff is pushing the boundaries on the "realistic" part.
There are occasional monster tits and sometimes a monster cock here. But the only major theme that connects all the different styles is that we depict roughly normal looking people (and occasionally a fairy).
A beautiful woman with a penis is just an inch away from strapon femdom and wouldn't bother me that much
But horns and tails creatures are a bit too far

File: 00520.jpg (820 KB, 3552x3904)

820 KB JPG

A revisit of one from the last thread with everything I've been learning. Wet hair is very tricky to prompt and look decent

>>8334573
looks great
you messed up in the upscaling though
looking at that pic at 100% zoom and it's blurry

File: Kaya_f0123.jpg (40 KB, 512x768)

40 KB JPG

>>8334559
Dude, take it to any other futa thread already on this board.

>>8334628
>barely recognizable kaya
>barely realistic, looks drawn
you are the expert on what belongs here!

>>8334603
Yeah, I've been kind of using the slight blur to cover up some masking and AI jank in the detail. I'll keep playing with it.

>>8334628
I like this style, but I don't know who Kaya is.

File: 00523.jpg (1.63 MB, 3552x3904)

1.63 MB JPG

>>8334603
Forgot to post one with a much sharper upscale

>>8332181
Shut up retard

>>8332788
Request goes on request board retard

File: 22104-322056438.png (1.67 MB, 1200x1600)

1.67 MB PNG

>>8334469
This isn't /b/, if people post furry here you can just report it.

>>8334461
Funny that AOC fag tried to vote on no futa

AOC is turbo retard left who would vote to turn all men into futa irl

>>8334649
epic
you sure you're new to this
>>8334666
this is the right answer. there is no slippery slope

File: 00264-3005409469-0000.jpg (502 KB, 3200x3200)

502 KB JPG

>>8334667
My brother in Christ, you seem to be suffering from political brainrot. AOC would not appreciate me creating fake images of her to jack off to and share them, who cares what she thinks?

>>8334707
I'm familiar with photo editing and had been playing around with some other AI stuff, but believe me my early gens were pretty rough

>>8334748
What can be done with inpainting now in minutes took me literally hours in GIMP before. And the technology is still in its infancy when you think about it.

That guy who calls people names is off his meds. Never provides any arguments. Only tries to discourage people from genning. And thinks that throwing insults around on an anonymous image board does anything. Just some pathetic nogen loser.

File: image_2024-06-22_112836939.png (1.18 MB, 832x1088)

1.18 MB PNG

>>8334212
You must learn to accept Dollar Store Daisy into your heart.

File: 00526.jpg (1.63 MB, 3229x4000)

1.63 MB JPG

>You interrupted my workout again... But your form is good, maybe you can spot me next time

We're dialing it in

>>8334894
maybe a sports bra can help

File: image_2024-06-22_123005611.png (919 KB, 960x768)

919 KB PNG

QUAD JUMPAH

>>8335164
Daisy head looks bigger compared to her body on all pictures but Carie Fischer is very good. Can you do this with Emila Clarke

>>8334748
But you're not making any AOC fakes

Just some dog faced latina

File: 00464-2835947367.png (1023 KB, 1024x1024)

1023 KB PNG

giwtwm

>>8335378
botched upscale again

>>8332706
Just checked the archive. Are you fucking retarded?

>>8335403
what's the argument? "CP was always posted here, so it's ok"?
the thread is for realistic gens. if you post pictures of naked kids, kindly rope yourself

File: 00532.jpg (1.27 MB, 3200x4000)

1.27 MB JPG

>>8334957
She sweats through everything anyway

So the troll who calls people retard in every single post is a kid fucker
Who is surprised

>>8332706
Well there are no real children so its not CP its just weird

>>8335535
AI generated pornography is pornography. So when it has kids in it, well then guess what kind of pornography that is?
No gray area like for anime since it's generated to look photo realistic.

>>8335545
you are gay

>>8335545
There are not real people or children involved in CGI, technically its not pornography, you are not taking videos or pics of real people its just weird CGI , those in the pics doesn't exist in real life

File: 22127-1674774228.png (1.63 MB, 1200x1600)

1.63 MB PNG

>>8335545
>>8335563
really real true facts incoming:
it's considered "obscenity" (and therefore not covered by the first amendment) if it's
>devoid of artistic merit
>appeals to the purient interest (read: makes you horny)
and it's considered CP if it's indistinguishable from the real thing to the average "reasonable person". And a quick scroll through *any* political debate on the internet should tell you that the average "reasonable person" in 2024 is a goddamn retard.

tl;dr: gen whatever you want but maybe try not to be a fucking idiot about what you post on the largest english-language imageboard on the entire web.

File: 00023-2689117003.png (1.06 MB, 992x1328)

1.06 MB PNG

>>8334007
some more i made today

File: 00027-2929068420.png (1.43 MB, 1560x904)

1.43 MB PNG

>>8335657

File: 00053-67481989.png (1.77 MB, 1568x1120)

1.77 MB PNG

>>8335659

File: 00045-829613738.png (1.15 MB, 752x1360)

1.15 MB PNG

>>8335660

>>8335453
1. that wasn't a loli
2. it wasn't realistic, it was cg at best
you are retarded

File: aco.jpg (127 KB, 800x1411)

127 KB JPG

hitting a cosmic ideal of "Zendaya or at least close enough", "beehive hair", a retro-ish kitchen, and "fingers aren't visible". Feel free to complain that I intentionally made her tits too big.

>>8335563
>>8335560
You people are idiots. This guy has it right >>8335628
Fucking delusional pedos.

im going to be 100% honest. this thread is amazing as fuck and the dalle thread have been dead all year. this will be THE thread until its not

>all these famous women's gens
>no Greta.

File: our_queen1.jpg (1.13 MB, 1139x1440)

1.13 MB JPG

>>8336224
Never heard of her.

>>8335659
>>8335660
>>8335661
Those are amazing. Make sure to show these to >>8331732

>>8336082
always point out what you like
there are too many amazing gens without a single reply
the genners need to know what's popular

>>8336224
as this thread gets more popular, we are going to have to have request anchors

>>8336246
/pol/ says hi! ;^]

>>8336284
fair point - there's probably a gen bred on /r/ but I just thought it was worth a mention, if we can have aoc, daisy and Emma why not Greta?
At least I'm not the cunt who kept asking for Emilia Clarke.

File: NSFW_A-00004.jpg (45 KB, 512x768)

45 KB JPG

>>8336290
Because that ugly swedish muppet is a magnet for poltards, so take that back to /b/. This tread is already turning into a shitfest.

File: 00043-1367141579.jpg (1.56 MB, 2240x3040)

1.56 MB JPG

tentacle gens are interesting. extremely hard to find one without little errors, and it seems to only want to use large bolt-on tits for things like "tentacle boob squeeze"

File: 00041-3984857896.jpg (1.72 MB, 2272x3040)

1.72 MB JPG

once again, not a banana

>>8336529
Interesting results non the less, bit more worrisome the fail with banana.

File: 00544.jpg (732 KB, 3103x4000)

732 KB JPG

It's a beautiful morning to gen

>>8336536
>>8336539
>>8336541
I really like the skin texture here

>>8336739
For this kind of room did you prompt more for "dark room" or "mood light" because I can't get rooms dark enough

>>8336536
>>8336537
>>8336539
>>8336541
you got your face mojo back -- no more dead doll's eyes
bravo

>>8336850
NTA but haven't found a silver bullet for dark shots yet
I think the problem is that photography style prompts are implying a well lit shot
darkest I got yet were by using "dark" adjective when describing the room
some strange results with "ethereal lighting, soft and otherworldly light"
would probably help to feed dark room shots into the CLIP interrogator to learn the vocabulary that SD associates with the desired lighting

>>8328883
>>8334006
>>8336445
You should try making Maisie, Greta or some other goblina
All your celebs look like goblins anyways, you should excel at these

>>8335485
You've got people mixed up or you're just projecting

Fucking retard

>>8337387
just make tripcode already
it's easier than typing your signature each time

File: 00535.jpg (788 KB, 4000x4000)

788 KB JPG

>Look, I'm all about inclusive, diverse spaces... But whenever you're here I can feel you staring at me the whole time

File: 00537.jpg (2.06 MB, 4000x4000)

2.06 MB JPG

>Stop drooling for a second and help me keep my back straight here

File: 00538.jpg (1.41 MB, 4000x4000)

1.41 MB JPG

>How does this look? Think I can go deeper?

File: image_e6pgugg9_1719171530(...).jpg (650 KB, 1024x992)

650 KB JPG

new to this hope u like

>>8337391
Fuck off retard

File: pic202.png (3 MB, 1024x1024)

3 MB PNG

>>8337464
another, i have alot of older ones i made from dalle, still getting used to this new one

>>8337470
I like this one, is it SD or DALL-E?

File: Designer - 2024-01-23 23_35_12.png (3 MB, 1024x1024)

3 MB PNG

>>8337475
dalle before it the whole taylor swift fag ruining it. i have tons more if you want, i know this isnt the right thread though

File: 00547.jpg (1.31 MB, 3552x4000)

1.31 MB JPG

>>8337482
You don't need to dump it if you don't want, I just want to emulate that dirty film look.

I don't know what the hell kind of non-euclidean geometry is going on, I guess the backseat of a car is harder than it seems

>>8337482
if you get a VPN to china you can get uncensored dalle again right now if you really want (cici)

>>8337507
yeah SD has trouble with tight spaces (or rather, the checkpoints have)
car interior, trunk, elevators... sometimes there is a LoRA available (like for elevators)
needs img2img/controlnet if you really want it
of course with that technique you can do almost anything anyways

File: pic128.png (3 MB, 1024x1024)

3 MB PNG

more nat

>>8337690
This is Bing dall-e you idiot

File: pic29.png (3 MB, 1024x1024)

3 MB PNG

8337482

already said that. posting more nat anyway now..

>>8337733
don't do a dump of dalle generated stuff here
that's not allowed (it's in the OP come on)
the dalle thread would love to reminisce about those I'm sure

>>8337733
more nat

>>8337733
>>8337837
over there >>8337276

File: image - 2024-06-24T001141.268.png (1.24 MB, 1152x896)

1.24 MB PNG

need a good sdxl/pony cleave gag lora or something. holes not stuffed enough

>>8333966
>>8333981
>>8333985
>>8333990
>>8333995
Nice gens, keep them

>>8338652
*coming, what models and quality tags are u using? Your gens look a bit much like a 3d render

File: 00154-1271084406.png (1.33 MB, 1080x1216)

1.33 MB PNG

>>8338348
like this?

File: 24175204944-6a385b.jpg (281 KB, 768x1344)

281 KB JPG

>>8336739
Not bad looking, but you forgot to set a photography style
It's basically a photo realistic render or painting and the realistic checkpoints somehow default to this
Can get some really interesting results if you take it further in that direction

>>8338396
plaplpalaplap get.... pregnant?

File: render-vs-photo.jpg (863 KB, 1536x1344)

863 KB JPG

>>8338736
Comparison example render vs. photo style
Also fuck Fooocus for including so many shit tags in their style prompts
Prompting is hard enough to learn and Fooocus is supposed to be easy for beginners, but boy they seem to have no idea what they're doing

>>8338736
>>8338823
That pic was from archive, but the point you make it´s extremely important. Tokens like 8k probably are superstition, but others will make small, subte yet impactful changes .

>>8337733
Bro this is good quality content, it has no place here

>>8338835
I think 8k actually does something to increase details, but haven't checked closely
Combinatorial generation gives lots of clues but it's tiresome
Shit like "vintage, shot on kodak" alters body types, fucking what (Fooocus Photography)
50mm, 75mm lenses and other camera stuff is supposed to make people look different (especially faces), but that shit goes too far

>>8338836
Really surprised how good dalle used to be, now there's cici at least
But don't be a smart ass and take it to the right thread
"dalle bros" always lose their shit whenever SD is even mentioned, so you can kindly fuck right off

>>8338844
8k I think is mosty cargo cult, doubt there was enough of that in the datasets to affect.
On the other hand vintage and shot on kodak probably did, moving the balance more to point and click, not-that-well thought out photography, that would clash with the more refined profesional compositions we want. And specifing lenses probably goes full on the other way.

>>8338860
Sometimes you can't tell a difference since what the tag you're adding would provide is already specified by other tags or LoRA anyways
Prompt without any style: add "8k" and you see big improvement
Prompt with photo/realistic style: adding "8k" won't change much
Unfortunately we cannot search the dataset anymore to see what was tagged in which way. They took all that down due to including problematic images by accident I think.
So it's all trial and error. And since the diffusion process is random, there is confirmation bias and cargo culting to worry about

>>8338883
Yeah. Concurring on all points.

just use prompt S/R on XYZ to see if "8k" have any effects on the image

Can AI generate a circumcised cock or are they all going to be uncut?

>>8338930
there are some blowjob POV LoRAs for XL on civitai that look like wrapped sausages

>>8338930
have you tried "shriveled glans", "scarred penis" or "jewish man"?

>>8330154
Mind sharing an example of an input image? Was wondering how many polys you are using/if any textures. Seems like a sick workflow for control but time consuming

>>8336536
Mmm is this Zonkey? Always struggle with detailed skin so catbox would be much appreciated

File: 00554.jpg (1.85 MB, 4000x4000)

1.85 MB JPG

>>8338883
Yeah I've tried copy+pasting prompts directly from Civitai and I get noticeably more generic outputs than me just using a few prompts, using ones without loras of course. I still put 8K in sometimes, but I have not noticed any real difference.

>>8338736
NTA, what do you mean photography style? Like just include that in the prompt?

>>8339082
Yeah so SD always needs to know if it should make a photo, a render, painting, drawing, concept art, etc.
That knowledge has to come from somewhere. The checkpoint (base) model might imply a default style. "Realistic" models make really good photos and make them by default.
Some part of your prompt should be dedicated to lock down a fixed style however. Those snippets you would reuse on many gens, so they are called "styles" and there are extensions to manage those also.
There are many different ways that photos can look. Camera lens (think tele lens vs. fish eye). Ever heard the phrase "the camera adds 10 pounds"?
Color balance, ISO/f-stop (hard or soft shadows), bokeh, depth of field etc.
You don't have to worry about all the details, you can for example just add "photographed by ..." or "in the style of ..." (famous photographer name there). Here is a list of known styles that work: https://rikkar69.github.io/SDXL-artist-study/tags/
You can filter by Photography tag on that page and try out what works.

File: 00552.jpg (1.07 MB, 3616x3616)

1.07 MB JPG

>>8339090
I've been prompting "Candid photo" but wasn't sure if it was actually doing anything

>>8339106
That does a lot, just including "photo". "Candid photography" can make people not look at the camera etc., the opposite of posing for the camera. But other parts of the prompt might override that.
If you are using a celeb lora, that is usually trained on photos only so that also tells SD to preferably make photos.
That's also why some of these LoRAs feel "stiff". If they got trained on all red carpet pictures and you don't prompt for pose and facial expression, you always get the same face and doll like posture.

File: dffd.jpg (148 KB, 1024x1024)

148 KB JPG

Emma-tentacles

File: 00010-3232607282.jpg (434 KB, 768x1344)

434 KB JPG

>>8339165
Ah, the Emma Watson Stylizer LoRA. Got deleted from civitai again
I don't know what to do with it except tentacles

File: 00555.jpg (1.48 MB, 3200x3616)

1.48 MB JPG

Wow, just adding a new photo lora and text2img is really cooking with gas with no extra work

>>8339026
that is pony realism… the main loras I’m using are “freckles and skin xl” at 1.0-1.4, the pony realism amateur style at 0.2 (forget the exact name but it’s linked on the pony realism civitai page) and add-detail-xl at 0.5-1.0

>>8339187
“We’d like you all to welcome Miss Hermione Granger back to Hogwarts - she will be teaching our new ‘Muggle Sexuality’ elective, which unsurprisingly has had a record number of applicants. I understand Miss Granger will be selecting her students personally in her bedr… er, office over the next several ni… er, days.”

File: 00558.jpg (908 KB, 2880x3520)

908 KB JPG

Now these feel like real candid shots

>>8339187
Can you make her with very hairy forearms and happy trail please?

>>8339412
I didn't realize how many XL LoRAs work so well with with realistic pony models then

>>8339431
Other stuff to try: amateur (subtly changes the scene sometimes), shot from below, shot from above, extreme angle
Now that you have turned up the skin detail she looks older. To go back try: cute, young and cute, insanely cute, unusually attractive, sexy young woman, attractive young woman
Ages also work: 20-year-old, teenager etc.
>>8339441
That is next level tit detail. Model/LoRA?

File: 24176205647-3388dc.jpg (398 KB, 768x1344)

398 KB JPG

I don't get how all the people on civitai making celeb LoRAs are so bad at making any good images with their creations
I will have to try out all the LoRAs, even the ones with only total garbage images to show for it
Making LoRAs is apparently easier than making images
Same face should be a bigger concern than getting good likeness

File: 00020-413561167.jpg (434 KB, 768x1344)

434 KB JPG

>>8339440
hard to get her bare arms to show when tentacles pull them back like Harvey Weinstein

File: 00557.jpg (1.03 MB, 2880x3520)

1.03 MB JPG

>>8339477
I just added the Steve McCurry lora at 1 and I also use the Intricate Details at .3~.5 but I'm not sure if that effects skin details. I got all of these from Cyberrealistic 4.2

>>8338844
Your reading comprehension is dogshit I was telling them to go back to their own thread

You gonna ever actually make her look like AOC instead of a filthy street rat

File: 00564.jpg (776 KB, 4000x3095)

776 KB JPG

Why should I do that?

>>8339606
>Bro this is good quality content, it has no place here
That's smart ass sarcasm
You are implying that all content here is low quality and dalle content is high quality
If you like it so much better, feel free to fuck off any time

File: 24176234946-df115b.jpg (398 KB, 768x1344)

398 KB JPG

>>8339706
Hey little sis, you look great

>>8339480
True I have to try all the loras really, because the examples are often really bad lol

Which model did you use for this one?

File: 24176200554-82243c.jpg (493 KB, 768x1344)

493 KB JPG

>>8339875
The SD1.5 lycoris by frankyfrank2k. It's only 70 MB.
That one actually has some good images except the forehead is always off. I fixed that by prompting for hair with bangs.
What I'm genning with her makes a younger version than those example pics somehow. I then supported that with "young and cute,no makeup".
It doesn't remove makeup but usually results in a younger look. All on Cyberrealistic checkpoint, but any good realistic model should work.
The XL Lora I tried briefly. It makes an older version of her. The MILF variant is a totally different beast that I'll tackle later.
Gonna grab the other SD1.5 LoRAs just in case. There is a rumor that civitai might go down the drain

File: image - 2024-06-24T122417.359.png (1.22 MB, 768x1344)

1.22 MB PNG

>>8339187
>>8339165

for inspiration

File: image - 2024-06-24T121555.420.png (1.24 MB, 1024x1024)

1.24 MB PNG

>>8339919
make sure you use refiner (with a decent XL model like epicrealism) and adetailer also with that checkpoint